Java tutorial
/** * Licensed to the Apache Software Foundation (ASF) under one * or more contributor license agreements. See the NOTICE file * distributed with this work for additional information * regarding copyright ownership. The ASF licenses this file * to you under the Apache License, Version 2.0 (the * "License"); you may not use this file except in compliance * with the License. You may obtain a copy of the License at * * http://www.apache.org/licenses/LICENSE-2.0 * * Unless required by applicable law or agreed to in writing, software * distributed under the License is distributed on an "AS IS" BASIS, * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. * See the License for the specific language governing permissions and * limitations under the License. */ package org.apache.hadoop.hive.metastore.txn; import com.google.common.annotations.VisibleForTesting; import com.jolbox.bonecp.BoneCPConfig; import com.jolbox.bonecp.BoneCPDataSource; import com.zaxxer.hikari.HikariConfig; import com.zaxxer.hikari.HikariDataSource; import org.apache.commons.dbcp.ConnectionFactory; import org.apache.commons.dbcp.DriverManagerConnectionFactory; import org.apache.commons.dbcp.PoolableConnectionFactory; import org.apache.commons.lang.NotImplementedException; import org.apache.hadoop.hive.common.ServerUtils; import org.apache.hadoop.hive.common.classification.InterfaceAudience; import org.apache.hadoop.hive.common.classification.InterfaceStability; import org.apache.hadoop.hive.common.classification.RetrySemantics; import org.apache.hadoop.hive.metastore.DatabaseProduct; import org.apache.hadoop.hive.metastore.HouseKeeperService; import org.apache.hadoop.hive.metastore.Warehouse; import org.slf4j.Logger; import org.slf4j.LoggerFactory; import org.apache.commons.dbcp.PoolingDataSource; import org.apache.commons.pool.ObjectPool; import org.apache.commons.pool.impl.GenericObjectPool; import org.apache.hadoop.hive.common.JavaUtils; import org.apache.hadoop.hive.common.StringableMap; import org.apache.hadoop.hive.conf.HiveConf; import org.apache.hadoop.hive.conf.HiveConfUtil; import org.apache.hadoop.hive.metastore.api.*; import org.apache.hadoop.hive.shims.ShimLoader; import org.apache.hadoop.util.StringUtils; import javax.sql.DataSource; import java.io.IOException; import java.io.PrintWriter; import java.nio.ByteBuffer; import java.sql.*; import java.util.*; import java.util.concurrent.ConcurrentHashMap; import java.util.concurrent.Semaphore; import java.util.concurrent.TimeUnit; import java.util.concurrent.locks.ReentrantLock; import java.util.regex.Pattern; /** * A handler to answer transaction related calls that come into the metastore * server. * * Note on log messages: Please include txnid:X and lockid info using * {@link org.apache.hadoop.hive.common.JavaUtils#txnIdToString(long)} * and {@link org.apache.hadoop.hive.common.JavaUtils#lockIdToString(long)} in all messages. * The txnid:X and lockid:Y matches how Thrift object toString() methods are generated, * so keeping the format consistent makes grep'ing the logs much easier. * * Note on HIVE_LOCKS.hl_last_heartbeat. * For locks that are part of transaction, we set this 0 (would rather set it to NULL but * Currently the DB schema has this NOT NULL) and only update/read heartbeat from corresponding * transaction in TXNS. * * In general there can be multiple metastores where this logic can execute, thus the DB is * used to ensure proper mutexing of operations. * Select ... For Update (or equivalent: either MsSql with(updlock) or actual Update stmt) is * used to properly sequence operations. Most notably: * 1. various sequence IDs are generated with aid of this mutex * 2. ensuring that each (Hive) Transaction state is transitioned atomically. Transaction state * includes its actual state (Open, Aborted) as well as it's lock list/component list. Thus all * per transaction ops, either start by update/delete of the relevant TXNS row or do S4U on that row. * This allows almost all operations to run at READ_COMMITTED and minimizes DB deadlocks. * 3. checkLock() - this is mutexted entirely since we must ensure that while we check if some lock * can be granted, no other (strictly speaking "earlier") lock can change state. * * The exception to his is Derby which doesn't support proper S4U. Derby is always running embedded * (this is the only supported configuration for Derby) * in the same JVM as HiveMetaStoreHandler thus we use JVM wide lock to properly sequnce the operations. * * {@link #derbyLock} * If we ever decide to run remote Derby server, according to * https://db.apache.org/derby/docs/10.0/manuals/develop/develop78.html all transactions will be * seriazlied, so that would also work though has not been tested. * * General design note: * It's imperative that any operation on a txn (e.g. commit), ensure (atomically) that this txn is * still valid and active. In the code this is usually achieved at the same time the txn record * is locked for some operation. * * Note on retry logic: * Metastore has retry logic in both {@link org.apache.hadoop.hive.metastore.RetryingMetaStoreClient} * and {@link org.apache.hadoop.hive.metastore.RetryingHMSHandler}. The retry logic there is very * generic and is not aware whether the operations are idempotent or not. (This is separate from * retry logic here in TxnHander which can/does retry DB errors intelligently). The worst case is * when an op here issues a successful commit against the RDBMS but the calling stack doesn't * receive the ack and retries. (If an op fails before commit, it's trivially idempotent) * Thus the ops here need to be made idempotent as much as possible or * the metstore call stack should have logic not to retry. There are {@link RetrySemantics} * annotations to document the behavior. */ @InterfaceAudience.Private @InterfaceStability.Evolving abstract class TxnHandler implements TxnStore, TxnStore.MutexAPI { static final protected char INITIATED_STATE = 'i'; static final protected char WORKING_STATE = 'w'; static final protected char READY_FOR_CLEANING = 'r'; static final char FAILED_STATE = 'f'; static final char SUCCEEDED_STATE = 's'; static final char ATTEMPTED_STATE = 'a'; // Compactor types static final protected char MAJOR_TYPE = 'a'; static final protected char MINOR_TYPE = 'i'; // Transaction states static final protected char TXN_ABORTED = 'a'; static final protected char TXN_OPEN = 'o'; //todo: make these like OperationType and remove above char constatns enum TxnStatus { OPEN, ABORTED, COMMITTED, UNKNOWN } // Lock states static final protected char LOCK_ACQUIRED = 'a'; static final protected char LOCK_WAITING = 'w'; // Lock types static final protected char LOCK_EXCLUSIVE = 'e'; static final protected char LOCK_SHARED = 'r'; static final protected char LOCK_SEMI_SHARED = 'w'; static final private int ALLOWED_REPEATED_DEADLOCKS = 10; static final private Logger LOG = LoggerFactory.getLogger(TxnHandler.class.getName()); static private DataSource connPool; private static DataSource connPoolMutex; static private boolean doRetryOnConnPool = false; private enum OpertaionType { SELECT('s'), INSERT('i'), UPDATE('u'), DELETE('d'); private final char sqlConst; OpertaionType(char sqlConst) { this.sqlConst = sqlConst; } public String toString() { return Character.toString(sqlConst); } public static OpertaionType fromString(char sqlConst) { switch (sqlConst) { case 's': return SELECT; case 'i': return INSERT; case 'u': return UPDATE; case 'd': return DELETE; default: throw new IllegalArgumentException(quoteChar(sqlConst)); } } public static OpertaionType fromDataOperationType(DataOperationType dop) { switch (dop) { case SELECT: return OpertaionType.SELECT; case INSERT: return OpertaionType.INSERT; case UPDATE: return OpertaionType.UPDATE; case DELETE: return OpertaionType.DELETE; default: throw new IllegalArgumentException("Unexpected value: " + dop); } } } // Maximum number of open transactions that's allowed private static volatile int maxOpenTxns = 0; // Current number of open txns private static volatile long numOpenTxns = 0; // Whether number of open transactions reaches the threshold private static volatile boolean tooManyOpenTxns = false; // The AcidHouseKeeperService for counting open transactions private static volatile HouseKeeperService openTxnsCounter = null; /** * Number of consecutive deadlocks we have seen */ private int deadlockCnt; private long deadlockRetryInterval; protected HiveConf conf; private static DatabaseProduct dbProduct; private static SQLGenerator sqlGenerator; // (End user) Transaction timeout, in milliseconds. private long timeout; private String identifierQuoteString; // quotes to use for quoting tables, where necessary private long retryInterval; private int retryLimit; private int retryNum; /** * Derby specific concurrency control */ private static final ReentrantLock derbyLock = new ReentrantLock(true); /** * must be static since even in UT there may be > 1 instance of TxnHandler * (e.g. via Compactor services) */ private final static ConcurrentHashMap<String, Semaphore> derbyKey2Lock = new ConcurrentHashMap<>(); private static final String hostname = ServerUtils.hostname(); // Private methods should never catch SQLException and then throw MetaException. The public // methods depend on SQLException coming back so they can detect and handle deadlocks. Private // methods should only throw MetaException when they explicitly know there's a logic error and // they want to throw past the public methods. // // All public methods that write to the database have to check for deadlocks when a SQLException // comes back and handle it if they see one. This has to be done with the connection pooling // in mind. To do this they should call checkRetryable() AFTER rolling back the db transaction, // and then they should catch RetryException and call themselves recursively. See commitTxn for an example. public TxnHandler() { } /** * This is logically part of c'tor and must be called prior to any other method. * Not physically part of c'tor due to use of relfection */ public void setConf(HiveConf conf) { this.conf = conf; checkQFileTestHack(); synchronized (TxnHandler.class) { if (connPool == null) { //only do this once per JVM; useful for support LOG.info(HiveConfUtil.dumpConfig(conf).toString()); Connection dbConn = null; // Set up the JDBC connection pool try { int maxPoolSize = conf .getIntVar(HiveConf.ConfVars.METASTORE_CONNECTION_POOLING_MAX_CONNECTIONS); long getConnectionTimeoutMs = 30000; connPool = setupJdbcConnectionPool(conf, maxPoolSize, getConnectionTimeoutMs); /*the mutex pools should ideally be somewhat larger since some operations require 1 connection from each pool and we want to avoid taking a connection from primary pool and then blocking because mutex pool is empty. There is only 1 thread in any HMS trying to mutex on each MUTEX_KEY except MUTEX_KEY.CheckLock. The CheckLock operation gets a connection from connPool first, then connPoolMutex. All others, go in the opposite order (not very elegant...). So number of connection requests for connPoolMutex cannot exceed (size of connPool + MUTEX_KEY.values().length - 1).*/ connPoolMutex = setupJdbcConnectionPool(conf, maxPoolSize + MUTEX_KEY.values().length, getConnectionTimeoutMs); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); determineDatabaseProduct(dbConn); sqlGenerator = new SQLGenerator(dbProduct, conf); } catch (SQLException e) { String msg = "Unable to instantiate JDBC connection pooling, " + e.getMessage(); LOG.error(msg); throw new RuntimeException(e); } finally { closeDbConn(dbConn); } } } timeout = HiveConf.getTimeVar(conf, HiveConf.ConfVars.HIVE_TXN_TIMEOUT, TimeUnit.MILLISECONDS); buildJumpTable(); retryInterval = HiveConf.getTimeVar(conf, HiveConf.ConfVars.HMSHANDLERINTERVAL, TimeUnit.MILLISECONDS); retryLimit = HiveConf.getIntVar(conf, HiveConf.ConfVars.HMSHANDLERATTEMPTS); deadlockRetryInterval = retryInterval / 10; maxOpenTxns = HiveConf.getIntVar(conf, HiveConf.ConfVars.HIVE_MAX_OPEN_TXNS); } @Override @RetrySemantics.ReadOnly public GetOpenTxnsInfoResponse getOpenTxnsInfo() throws MetaException { try { // We need to figure out the current transaction number and the list of // open transactions. To avoid needing a transaction on the underlying // database we'll look at the current transaction number first. If it // subsequently shows up in the open list that's ok. Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { /** * This method can run at READ_COMMITTED as long as long as * {@link #openTxns(org.apache.hadoop.hive.metastore.api.OpenTxnRequest)} is atomic. * More specifically, as long as advancing TransactionID in NEXT_TXN_ID is atomic with * adding corresponding entries into TXNS. The reason is that any txnid below HWM * is either in TXNS and thus considered open (Open/Aborted) or it's considered Committed. */ dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select ntxn_next - 1 from NEXT_TXN_ID"; LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { throw new MetaException( "Transaction tables not properly " + "initialized, no record found in next_txn_id"); } long hwm = rs.getLong(1); if (rs.wasNull()) { throw new MetaException( "Transaction tables not properly " + "initialized, null record found in next_txn_id"); } close(rs); List<TxnInfo> txnInfos = new ArrayList<TxnInfo>(); //need the WHERE clause below to ensure consistent results with READ_COMMITTED s = "select txn_id, txn_state, txn_user, txn_host, txn_started, txn_last_heartbeat from " + "TXNS where txn_id <= " + hwm; LOG.debug("Going to execute query<" + s + ">"); rs = stmt.executeQuery(s); while (rs.next()) { char c = rs.getString(2).charAt(0); TxnState state; switch (c) { case TXN_ABORTED: state = TxnState.ABORTED; break; case TXN_OPEN: state = TxnState.OPEN; break; default: throw new MetaException("Unexpected transaction state " + c + " found in txns table"); } TxnInfo txnInfo = new TxnInfo(rs.getLong(1), state, rs.getString(3), rs.getString(4)); txnInfo.setStartedTime(rs.getLong(5)); txnInfo.setLastHeartbeatTime(rs.getLong(6)); txnInfos.add(txnInfo); } LOG.debug("Going to rollback"); dbConn.rollback(); return new GetOpenTxnsInfoResponse(hwm, txnInfos); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "getOpenTxnsInfo"); throw new MetaException("Unable to select from transaction database: " + getMessage(e) + StringUtils.stringifyException(e)); } finally { close(rs, stmt, dbConn); } } catch (RetryException e) { return getOpenTxnsInfo(); } } @Override @RetrySemantics.ReadOnly public GetOpenTxnsResponse getOpenTxns() throws MetaException { try { // We need to figure out the current transaction number and the list of // open transactions. To avoid needing a transaction on the underlying // database we'll look at the current transaction number first. If it // subsequently shows up in the open list that's ok. Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { /** * This runs at READ_COMMITTED for exactly the same reason as {@link #getOpenTxnsInfo()} */ dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select ntxn_next - 1 from NEXT_TXN_ID"; LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { throw new MetaException( "Transaction tables not properly " + "initialized, no record found in next_txn_id"); } long hwm = rs.getLong(1); if (rs.wasNull()) { throw new MetaException( "Transaction tables not properly " + "initialized, null record found in next_txn_id"); } close(rs); List<Long> openList = new ArrayList<Long>(); //need the WHERE clause below to ensure consistent results with READ_COMMITTED s = "select txn_id, txn_state from TXNS where txn_id <= " + hwm + " order by txn_id"; LOG.debug("Going to execute query<" + s + ">"); rs = stmt.executeQuery(s); long minOpenTxn = Long.MAX_VALUE; BitSet abortedBits = new BitSet(); while (rs.next()) { long txnId = rs.getLong(1); openList.add(txnId); char c = rs.getString(2).charAt(0); if (c == TXN_OPEN) { minOpenTxn = Math.min(minOpenTxn, txnId); } else if (c == TXN_ABORTED) { abortedBits.set(openList.size() - 1); } } LOG.debug("Going to rollback"); dbConn.rollback(); ByteBuffer byteBuffer = ByteBuffer.wrap(abortedBits.toByteArray()); GetOpenTxnsResponse otr = new GetOpenTxnsResponse(hwm, openList, byteBuffer); if (minOpenTxn < Long.MAX_VALUE) { otr.setMin_open_txn(minOpenTxn); } return otr; } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "getOpenTxns"); throw new MetaException( "Unable to select from transaction database, " + StringUtils.stringifyException(e)); } finally { close(rs, stmt, dbConn); } } catch (RetryException e) { return getOpenTxns(); } } private static void startHouseKeeperService(HiveConf conf, Class c) { try { openTxnsCounter = (HouseKeeperService) c.newInstance(); openTxnsCounter.start(conf); } catch (Exception ex) { LOG.error("Failed to start {}", openTxnsCounter.getClass() + ". The system will not handle {} ", openTxnsCounter.getServiceDescription(), ". Root Cause: ", ex); } } /** * Retry-by-caller note: * Worst case, it will leave an open txn which will timeout. */ @Override @RetrySemantics.Idempotent public OpenTxnsResponse openTxns(OpenTxnRequest rqst) throws MetaException { if (openTxnsCounter == null) { synchronized (TxnHandler.class) { try { if (openTxnsCounter == null) { startHouseKeeperService(conf, Class.forName("org.apache.hadoop.hive.ql.txn.AcidOpenTxnsCounterService")); } } catch (ClassNotFoundException e) { throw new MetaException(e.getMessage()); } } } if (!tooManyOpenTxns && numOpenTxns >= maxOpenTxns) { tooManyOpenTxns = true; } if (tooManyOpenTxns) { if (numOpenTxns < maxOpenTxns * 0.9) { tooManyOpenTxns = false; } else { LOG.warn("Maximum allowed number of open transactions (" + maxOpenTxns + ") has been " + "reached. Current number of open transactions: " + numOpenTxns); throw new MetaException("Maximum allowed number of open transactions has been reached. " + "See hive.max.open.txns."); } } int numTxns = rqst.getNum_txns(); try { Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { lockInternal(); /** * To make {@link #getOpenTxns()}/{@link #getOpenTxnsInfo()} work correctly, this operation must ensure * that advancing the counter in NEXT_TXN_ID and adding appropriate entries to TXNS is atomic. * Also, advancing the counter must work when multiple metastores are running. * SELECT ... FOR UPDATE is used to prevent * concurrent DB transactions being rolled back due to Write-Write conflict on NEXT_TXN_ID. * * In the current design, there can be several metastore instances running in a given Warehouse. * This makes ideas like reserving a range of IDs to save trips to DB impossible. For example, * a client may go to MS1 and start a transaction with ID 500 to update a particular row. * Now the same client will start another transaction, except it ends up on MS2 and may get * transaction ID 400 and update the same row. Now the merge that happens to materialize the snapshot * on read will thing the version of the row from transaction ID 500 is the latest one. * * Longer term we can consider running Active-Passive MS (at least wrt to ACID operations). This * set could support a write-through cache for added performance. */ dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); // Make sure the user has not requested an insane amount of txns. int maxTxns = HiveConf.getIntVar(conf, HiveConf.ConfVars.HIVE_TXN_MAX_OPEN_BATCH); if (numTxns > maxTxns) numTxns = maxTxns; stmt = dbConn.createStatement(); String s = sqlGenerator.addForUpdateClause("select ntxn_next from NEXT_TXN_ID"); LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { throw new MetaException( "Transaction database not properly " + "configured, can't find next transaction id."); } long first = rs.getLong(1); s = "update NEXT_TXN_ID set ntxn_next = " + (first + numTxns); LOG.debug("Going to execute update <" + s + ">"); stmt.executeUpdate(s); long now = getDbTime(dbConn); List<Long> txnIds = new ArrayList<Long>(numTxns); List<String> rows = new ArrayList<>(); for (long i = first; i < first + numTxns; i++) { txnIds.add(i); rows.add(i + "," + quoteChar(TXN_OPEN) + "," + now + "," + now + "," + quoteString(rqst.getUser()) + "," + quoteString(rqst.getHostname())); } List<String> queries = sqlGenerator.createInsertValuesStmt( "TXNS (txn_id, txn_state, txn_started, txn_last_heartbeat, txn_user, txn_host)", rows); for (String q : queries) { LOG.debug("Going to execute update <" + q + ">"); stmt.execute(q); } LOG.debug("Going to commit"); dbConn.commit(); return new OpenTxnsResponse(txnIds); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "openTxns(" + rqst + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { close(rs, stmt, dbConn); unlockInternal(); } } catch (RetryException e) { return openTxns(rqst); } } @Override @RetrySemantics.Idempotent public void abortTxn(AbortTxnRequest rqst) throws NoSuchTxnException, MetaException, TxnAbortedException { long txnid = rqst.getTxnid(); try { Connection dbConn = null; Statement stmt = null; try { lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); if (abortTxns(dbConn, Collections.singletonList(txnid), true) != 1) { stmt = dbConn.createStatement(); TxnStatus status = findTxnState(txnid, stmt); if (status == TxnStatus.ABORTED) { LOG.info("abortTxn(" + JavaUtils.txnIdToString(txnid) + ") requested by it is already " + TxnStatus.ABORTED); return; } raiseTxnUnexpectedState(status, txnid); } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "abortTxn(" + rqst + ")"); throw new MetaException( "Unable to update transaction database " + StringUtils.stringifyException(e)); } finally { close(null, stmt, dbConn); unlockInternal(); } } catch (RetryException e) { abortTxn(rqst); } } @Override @RetrySemantics.Idempotent public void abortTxns(AbortTxnsRequest rqst) throws NoSuchTxnException, MetaException { List<Long> txnids = rqst.getTxn_ids(); try { Connection dbConn = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); int numAborted = abortTxns(dbConn, txnids, false); if (numAborted != txnids.size()) { LOG.warn("Abort Transactions command only aborted " + numAborted + " out of " + txnids.size() + " transactions. It's possible that the other " + (txnids.size() - numAborted) + " transactions have been aborted or committed, or the transaction ids are invalid."); } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "abortTxns(" + rqst + ")"); throw new MetaException( "Unable to update transaction database " + StringUtils.stringifyException(e)); } finally { closeDbConn(dbConn); } } catch (RetryException e) { abortTxns(rqst); } } /** * Concurrency/isolation notes: * This is mutexed with {@link #openTxns(OpenTxnRequest)} and other {@link #commitTxn(CommitTxnRequest)} * operations using select4update on NEXT_TXN_ID. Also, mutexes on TXNX table for specific txnid:X * see more notes below. * In order to prevent lost updates, we need to determine if any 2 transactions overlap. Each txn * is viewed as an interval [M,N]. M is the txnid and N is taken from the same NEXT_TXN_ID sequence * so that we can compare commit time of txn T with start time of txn S. This sequence can be thought of * as a logical time counter. If S.commitTime < T.startTime, T and S do NOT overlap. * * Motivating example: * Suppose we have multi-statment transactions T and S both of which are attempting x = x + 1 * In order to prevent lost update problem, the the non-overlapping txns must lock in the snapshot * that they read appropriately. In particular, if txns do not overlap, then one follows the other * (assumig they write the same entity), and thus the 2nd must see changes of the 1st. We ensure * this by locking in snapshot after * {@link #openTxns(OpenTxnRequest)} call is made (see {@link org.apache.hadoop.hive.ql.Driver#acquireLocksAndOpenTxn()}) * and mutexing openTxn() with commit(). In other words, once a S.commit() starts we must ensure * that txn T which will be considered a later txn, locks in a snapshot that includes the result * of S's commit (assuming no other txns). * As a counter example, suppose we have S[3,3] and T[4,4] (commitId=txnid means no other transactions * were running in parallel). If T and S both locked in the same snapshot (for example commit of * txnid:2, which is possible if commitTxn() and openTxnx() is not mutexed) * 'x' would be updated to the same value by both, i.e. lost update. */ @Override @RetrySemantics.Idempotent("No-op if already committed") public void commitTxn(CommitTxnRequest rqst) throws NoSuchTxnException, TxnAbortedException, MetaException { long txnid = rqst.getTxnid(); try { Connection dbConn = null; Statement stmt = null; ResultSet lockHandle = null; ResultSet commitIdRs = null, rs; try { lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); /** * Runs at READ_COMMITTED with S4U on TXNS row for "txnid". S4U ensures that no other * operation can change this txn (such acquiring locks). While lock() and commitTxn() * should not normally run concurrently (for same txn) but could due to bugs in the client * which could then corrupt internal transaction manager state. Also competes with abortTxn(). */ lockHandle = lockTransactionRecord(stmt, txnid, TXN_OPEN); if (lockHandle == null) { //if here, txn was not found (in expected state) TxnStatus actualTxnStatus = findTxnState(txnid, stmt); if (actualTxnStatus == TxnStatus.COMMITTED) { /** * This makes the operation idempotent * (assume that this is most likely due to retry logic) */ LOG.info("Nth commitTxn(" + JavaUtils.txnIdToString(txnid) + ") msg"); return; } raiseTxnUnexpectedState(actualTxnStatus, txnid); shouldNeverHappen(txnid); //dbConn is rolled back in finally{} } String conflictSQLSuffix = "from TXN_COMPONENTS where tc_txnid=" + txnid + " and tc_operation_type IN(" + quoteChar(OpertaionType.UPDATE.sqlConst) + "," + quoteChar(OpertaionType.DELETE.sqlConst) + ")"; rs = stmt.executeQuery(sqlGenerator.addLimitClause(1, "tc_operation_type " + conflictSQLSuffix)); if (rs.next()) { close(rs); //if here it means currently committing txn performed update/delete and we should check WW conflict /** * This S4U will mutex with other commitTxn() and openTxns(). * -1 below makes txn intervals look like [3,3] [4,4] if all txns are serial * Note: it's possible to have several txns have the same commit id. Suppose 3 txns start * at the same time and no new txns start until all 3 commit. * We could've incremented the sequence for commitId is well but it doesn't add anything functionally. */ commitIdRs = stmt .executeQuery(sqlGenerator.addForUpdateClause("select ntxn_next - 1 from NEXT_TXN_ID")); if (!commitIdRs.next()) { throw new IllegalStateException("No rows found in NEXT_TXN_ID"); } long commitId = commitIdRs.getLong(1); Savepoint undoWriteSetForCurrentTxn = dbConn.setSavepoint(); /** * "select distinct" is used below because * 1. once we get to multi-statement txns, we only care to record that something was updated once * 2. if {@link #addDynamicPartitions(AddDynamicPartitions)} is retried by caller it my create * duplicate entries in TXN_COMPONENTS * but we want to add a PK on WRITE_SET which won't have unique rows w/o this distinct * even if it includes all of it's columns */ int numCompsWritten = stmt.executeUpdate( "insert into WRITE_SET (ws_database, ws_table, ws_partition, ws_txnid, ws_commit_id, ws_operation_type)" + " select distinct tc_database, tc_table, tc_partition, tc_txnid, " + commitId + ", tc_operation_type " + conflictSQLSuffix); /** * see if there are any overlapping txns wrote the same element, i.e. have a conflict * Since entire commit operation is mutexed wrt other start/commit ops, * committed.ws_commit_id <= current.ws_commit_id for all txns * thus if committed.ws_commit_id < current.ws_txnid, transactions do NOT overlap * For example, [17,20] is committed, [6,80] is being committed right now - these overlap * [17,20] committed and [21,21] committing now - these do not overlap. * [17,18] committed and [18,19] committing now - these overlap (here 18 started while 17 was still running) */ rs = stmt.executeQuery(sqlGenerator.addLimitClause(1, "committed.ws_txnid, committed.ws_commit_id, committed.ws_database," + "committed.ws_table, committed.ws_partition, cur.ws_commit_id cur_ws_commit_id, " + "cur.ws_operation_type cur_op, committed.ws_operation_type committed_op " + "from WRITE_SET committed INNER JOIN WRITE_SET cur " + "ON committed.ws_database=cur.ws_database and committed.ws_table=cur.ws_table " + //For partitioned table we always track writes at partition level (never at table) //and for non partitioned - always at table level, thus the same table should never //have entries with partition key and w/o "and (committed.ws_partition=cur.ws_partition or (committed.ws_partition is null and cur.ws_partition is null)) " + "where cur.ws_txnid <= committed.ws_commit_id" + //txns overlap; could replace ws_txnid // with txnid, though any decent DB should infer this " and cur.ws_txnid=" + txnid + //make sure RHS of join only has rows we just inserted as // part of this commitTxn() op " and committed.ws_txnid <> " + txnid + //and LHS only has committed txns //U+U and U+D is a conflict but D+D is not and we don't currently track I in WRITE_SET at all " and (committed.ws_operation_type=" + quoteChar(OpertaionType.UPDATE.sqlConst) + " OR cur.ws_operation_type=" + quoteChar(OpertaionType.UPDATE.sqlConst) + ")")); if (rs.next()) { //found a conflict String committedTxn = "[" + JavaUtils.txnIdToString(rs.getLong(1)) + "," + rs.getLong(2) + "]"; StringBuilder resource = new StringBuilder(rs.getString(3)).append("/") .append(rs.getString(4)); String partitionName = rs.getString(5); if (partitionName != null) { resource.append('/').append(partitionName); } String msg = "Aborting [" + JavaUtils.txnIdToString(txnid) + "," + rs.getLong(6) + "]" + " due to a write conflict on " + resource + " committed by " + committedTxn + " " + rs.getString(7) + "/" + rs.getString(8); close(rs); //remove WRITE_SET info for current txn since it's about to abort dbConn.rollback(undoWriteSetForCurrentTxn); LOG.info(msg); //todo: should make abortTxns() write something into TXNS.TXN_META_INFO about this if (abortTxns(dbConn, Collections.singletonList(txnid), true) != 1) { throw new IllegalStateException(msg + " FAILED!"); } dbConn.commit(); close(null, stmt, dbConn); throw new TxnAbortedException(msg); } else { //no conflicting operations, proceed with the rest of commit sequence } } else { /** * current txn didn't update/delete anything (may have inserted), so just proceed with commit * * We only care about commit id for write txns, so for RO (when supported) txns we don't * have to mutex on NEXT_TXN_ID. * Consider: if RO txn is after a W txn, then RO's openTxns() will be mutexed with W's * commitTxn() because both do S4U on NEXT_TXN_ID and thus RO will see result of W txn. * If RO < W, then there is no reads-from relationship. */ } // Move the record from txn_components into completed_txn_components so that the compactor // knows where to look to compact. String s = "insert into COMPLETED_TXN_COMPONENTS select tc_txnid, tc_database, tc_table, " + "tc_partition from TXN_COMPONENTS where tc_txnid = " + txnid; LOG.debug("Going to execute insert <" + s + ">"); int modCount = 0; if ((modCount = stmt.executeUpdate(s)) < 1) { //this can be reasonable for an empty txn START/COMMIT or read-only txn //also an IUD with DP that didn't match any rows. LOG.info("Expected to move at least one record from txn_components to " + "completed_txn_components when committing txn! " + JavaUtils.txnIdToString(txnid)); } s = "delete from TXN_COMPONENTS where tc_txnid = " + txnid; LOG.debug("Going to execute update <" + s + ">"); modCount = stmt.executeUpdate(s); s = "delete from HIVE_LOCKS where hl_txnid = " + txnid; LOG.debug("Going to execute update <" + s + ">"); modCount = stmt.executeUpdate(s); s = "delete from TXNS where txn_id = " + txnid; LOG.debug("Going to execute update <" + s + ">"); modCount = stmt.executeUpdate(s); LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "commitTxn(" + rqst + ")"); throw new MetaException( "Unable to update transaction database " + StringUtils.stringifyException(e)); } finally { close(commitIdRs); close(lockHandle, stmt, dbConn); unlockInternal(); } } catch (RetryException e) { commitTxn(rqst); } } @Override @RetrySemantics.SafeToRetry public void performWriteSetGC() { Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); rs = stmt.executeQuery("select ntxn_next - 1 from NEXT_TXN_ID"); if (!rs.next()) { throw new IllegalStateException("NEXT_TXN_ID is empty: DB is corrupted"); } long highestAllocatedTxnId = rs.getLong(1); close(rs); rs = stmt.executeQuery("select min(txn_id) from TXNS where txn_state=" + quoteChar(TXN_OPEN)); if (!rs.next()) { throw new IllegalStateException("Scalar query returned no rows?!?!!"); } long commitHighWaterMark;//all currently open txns (if any) have txnid >= than commitHighWaterMark long lowestOpenTxnId = rs.getLong(1); if (rs.wasNull()) { //if here then there are no Open txns and highestAllocatedTxnId must be //resolved (i.e. committed or aborted), either way //there are no open txns with id <= highestAllocatedTxnId //the +1 is there because "delete ..." below has < (which is correct for the case when //there is an open txn //Concurrency: even if new txn starts (or starts + commits) it is still true that //there are no currently open txns that overlap with any committed txn with //commitId <= commitHighWaterMark (as set on next line). So plain READ_COMMITTED is enough. commitHighWaterMark = highestAllocatedTxnId + 1; } else { commitHighWaterMark = lowestOpenTxnId; } int delCnt = stmt.executeUpdate("delete from WRITE_SET where ws_commit_id < " + commitHighWaterMark); LOG.info("Deleted " + delCnt + " obsolete rows from WRTIE_SET"); dbConn.commit(); } catch (SQLException ex) { LOG.warn("WriteSet GC failed due to " + getMessage(ex), ex); } finally { close(rs, stmt, dbConn); } } /** * As much as possible (i.e. in absence of retries) we want both operations to be done on the same * connection (but separate transactions). This avoid some flakiness in BONECP where if you * perform an operation on 1 connection and immediately get another from the pool, the 2nd one * doesn't see results of the first. * * Retry-by-caller note: If the call to lock is from a transaction, then in the worst case * there will be a duplicate set of locks but both sets will belong to the same txn so they * will not conflict with each other. For locks w/o txn context (i.e. read-only query), this * may lead to deadlock (at least a long wait). (e.g. 1st call creates locks in {@code LOCK_WAITING} * mode and response gets lost. Then {@link org.apache.hadoop.hive.metastore.RetryingMetaStoreClient} * retries, and enqueues another set of locks in LOCK_WAITING. The 2nd LockResponse is delivered * to the DbLockManager, which will keep dong {@link #checkLock(CheckLockRequest)} until the 1st * set of locks times out. */ @RetrySemantics.CannotRetry public LockResponse lock(LockRequest rqst) throws NoSuchTxnException, TxnAbortedException, MetaException { ConnectionLockIdPair connAndLockId = enqueueLockWithRetry(rqst); try { return checkLockWithRetry(connAndLockId.dbConn, connAndLockId.extLockId, rqst.getTxnid()); } catch (NoSuchLockException e) { // This should never happen, as we just added the lock id throw new MetaException("Couldn't find a lock we just created! " + e.getMessage()); } } private static final class ConnectionLockIdPair { private final Connection dbConn; private final long extLockId; private ConnectionLockIdPair(Connection dbConn, long extLockId) { this.dbConn = dbConn; this.extLockId = extLockId; } } /** * Note that by definition select for update is divorced from update, i.e. you executeQuery() to read * and then executeUpdate(). One other alternative would be to actually update the row in TXNS but * to the same value as before thus forcing db to acquire write lock for duration of the transaction. * * There is no real reason to return the ResultSet here other than to make sure the reference to it * is retained for duration of intended lock scope and is not GC'd thus (unlikely) causing lock * to be released. * @param txnState the state this txn is expected to be in. may be null * @return null if no row was found * @throws SQLException * @throws MetaException */ private ResultSet lockTransactionRecord(Statement stmt, long txnId, Character txnState) throws SQLException, MetaException { String query = "select TXN_STATE from TXNS where TXN_ID = " + txnId + (txnState != null ? " AND TXN_STATE=" + quoteChar(txnState) : ""); ResultSet rs = stmt.executeQuery(sqlGenerator.addForUpdateClause(query)); if (rs.next()) { return rs; } close(rs); return null; } /** * This enters locks into the queue in {@link #LOCK_WAITING} mode. * * Isolation Level Notes: * 1. We use S4U (withe read_committed) to generate the next (ext) lock id. This serializes * any 2 {@code enqueueLockWithRetry()} calls. * 2. We use S4U on the relevant TXNS row to block any concurrent abort/commit/etc operations * @see #checkLockWithRetry(Connection, long, long) */ private ConnectionLockIdPair enqueueLockWithRetry(LockRequest rqst) throws NoSuchTxnException, TxnAbortedException, MetaException { boolean success = false; Connection dbConn = null; try { Statement stmt = null; ResultSet rs = null; ResultSet lockHandle = null; try { lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); long txnid = rqst.getTxnid(); stmt = dbConn.createStatement(); if (isValidTxn(txnid)) { //this also ensures that txn is still there in expected state lockHandle = lockTransactionRecord(stmt, txnid, TXN_OPEN); if (lockHandle == null) { ensureValidTxn(dbConn, txnid, stmt); shouldNeverHappen(txnid); } } /** Get the next lock id. * This has to be atomic with adding entries to HIVE_LOCK entries (1st add in W state) to prevent a race. * Suppose ID gen is a separate txn and 2 concurrent lock() methods are running. 1st one generates nl_next=7, * 2nd nl_next=8. Then 8 goes first to insert into HIVE_LOCKS and acquires the locks. Then 7 unblocks, * and add it's W locks but it won't see locks from 8 since to be 'fair' {@link #checkLock(java.sql.Connection, long)} * doesn't block on locks acquired later than one it's checking*/ String s = sqlGenerator.addForUpdateClause("select nl_next from NEXT_LOCK_ID"); LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { LOG.debug("Going to rollback"); dbConn.rollback(); throw new MetaException( "Transaction tables not properly " + "initialized, no record found in next_lock_id"); } long extLockId = rs.getLong(1); s = "update NEXT_LOCK_ID set nl_next = " + (extLockId + 1); LOG.debug("Going to execute update <" + s + ">"); stmt.executeUpdate(s); if (txnid > 0) { List<String> rows = new ArrayList<>(); // For each component in this lock request, // add an entry to the txn_components table for (LockComponent lc : rqst.getComponent()) { if (lc.isSetIsAcid() && !lc.isIsAcid()) { //we don't prevent using non-acid resources in a txn but we do lock them continue; } boolean updateTxnComponents; if (!lc.isSetOperationType()) { //request came from old version of the client updateTxnComponents = true;//this matches old behavior } else { switch (lc.getOperationType()) { case INSERT: case UPDATE: case DELETE: if (!lc.isSetIsDynamicPartitionWrite()) { //must be old client talking, i.e. we don't know if it's DP so be conservative updateTxnComponents = true; } else { /** * we know this is part of DP operation and so we'll get * {@link #addDynamicPartitions(AddDynamicPartitions)} call with the list * of partitions actually chaged. */ updateTxnComponents = !lc.isIsDynamicPartitionWrite(); } break; case SELECT: updateTxnComponents = false; break; case NO_TXN: /*this constant is a bit of a misnomer since we now always have a txn context. It just means the operation is such that we don't care what tables/partitions it affected as it doesn't trigger a compaction or conflict detection. A better name would be NON_TRANSACTIONAL.*/ updateTxnComponents = false; break; default: //since we have an open transaction, only 4 values above are expected throw new IllegalStateException( "Unexpected DataOperationType: " + lc.getOperationType() + " agentInfo=" + rqst.getAgentInfo() + " " + JavaUtils.txnIdToString(txnid)); } } if (!updateTxnComponents) { continue; } String dbName = lc.getDbname(); String tblName = lc.getTablename(); String partName = lc.getPartitionname(); rows.add(txnid + ", '" + dbName + "', " + (tblName == null ? "null" : "'" + tblName + "'") + ", " + (partName == null ? "null" : "'" + partName + "'") + "," + quoteString( OpertaionType.fromDataOperationType(lc.getOperationType()).toString())); } List<String> queries = sqlGenerator.createInsertValuesStmt( "TXN_COMPONENTS (tc_txnid, tc_database, tc_table, tc_partition, tc_operation_type)", rows); for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); int modCount = stmt.executeUpdate(query); } } List<String> rows = new ArrayList<>(); long intLockId = 0; for (LockComponent lc : rqst.getComponent()) { if (lc.isSetOperationType() && lc.getOperationType() == DataOperationType.UNSET && (conf.getBoolVar(HiveConf.ConfVars.HIVE_IN_TEST) || conf.getBoolVar(HiveConf.ConfVars.HIVE_IN_TEZ_TEST))) { //old version of thrift client should have (lc.isSetOperationType() == false) but they do not //If you add a default value to a variable, isSet() for that variable is true regardless of the where the //message was created (for object variables. It works correctly for boolean vars, e.g. LockComponent.isAcid). //in test mode, upgrades are not tested, so client version and server version of thrift always matches so //we see UNSET here it means something didn't set the appropriate value. throw new IllegalStateException("Bug: operationType=" + lc.getOperationType() + " for component " + lc + " agentInfo=" + rqst.getAgentInfo()); } intLockId++; String dbName = lc.getDbname(); String tblName = lc.getTablename(); String partName = lc.getPartitionname(); LockType lockType = lc.getType(); char lockChar = 'z'; switch (lockType) { case EXCLUSIVE: lockChar = LOCK_EXCLUSIVE; break; case SHARED_READ: lockChar = LOCK_SHARED; break; case SHARED_WRITE: lockChar = LOCK_SEMI_SHARED; break; } long now = getDbTime(dbConn); rows.add(extLockId + ", " + intLockId + "," + txnid + ", " + quoteString(dbName) + ", " + valueOrNullLiteral(tblName) + ", " + valueOrNullLiteral(partName) + ", " + quoteChar(LOCK_WAITING) + ", " + quoteChar(lockChar) + ", " + //for locks associated with a txn, we always heartbeat txn and timeout based on that (isValidTxn(txnid) ? 0 : now) + ", " + valueOrNullLiteral(rqst.getUser()) + ", " + valueOrNullLiteral(rqst.getHostname()) + ", " + valueOrNullLiteral(rqst.getAgentInfo()));// + ")"; } List<String> queries = sqlGenerator .createInsertValuesStmt("HIVE_LOCKS (hl_lock_ext_id, hl_lock_int_id, hl_txnid, hl_db, " + "hl_table, hl_partition,hl_lock_state, hl_lock_type, " + "hl_last_heartbeat, hl_user, hl_host, hl_agent_info)", rows); for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); int modCount = stmt.executeUpdate(query); } dbConn.commit(); success = true; return new ConnectionLockIdPair(dbConn, extLockId); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "enqueueLockWithRetry(" + rqst + ")"); throw new MetaException( "Unable to update transaction database " + StringUtils.stringifyException(e)); } finally { close(lockHandle); close(rs, stmt, null); if (!success) { /* This needs to return a "live" connection to be used by operation that follows it. Thus it only closes Connection on failure/retry. */ closeDbConn(dbConn); } unlockInternal(); } } catch (RetryException e) { return enqueueLockWithRetry(rqst); } } private LockResponse checkLockWithRetry(Connection dbConn, long extLockId, long txnId) throws NoSuchLockException, NoSuchTxnException, TxnAbortedException, MetaException { try { try { lockInternal(); if (dbConn.isClosed()) { //should only get here if retrying this op dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); } return checkLock(dbConn, extLockId); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "checkLockWithRetry(" + extLockId + "," + txnId + ")"); throw new MetaException( "Unable to update transaction database " + StringUtils.stringifyException(e)); } finally { unlockInternal(); closeDbConn(dbConn); } } catch (RetryException e) { return checkLockWithRetry(dbConn, extLockId, txnId); } } /** * Why doesn't this get a txnid as parameter? The caller should either know the txnid or know there isn't one. * Either way getTxnIdFromLockId() will not be needed. This would be a Thrift change. * * Also, when lock acquisition returns WAITING, it's retried every 15 seconds (best case, see DbLockManager.backoff(), * in practice more often) * which means this is heartbeating way more often than hive.txn.timeout and creating extra load on DB. * * The clients that operate in blocking mode, can't heartbeat a lock until the lock is acquired. * We should make CheckLockRequest include timestamp or last request to skip unnecessary heartbeats. Thrift change. * * {@link #checkLock(java.sql.Connection, long)} must run at SERIALIZABLE (make sure some lock we are checking * against doesn't move from W to A in another txn) but this method can heartbeat in * separate txn at READ_COMMITTED. * * Retry-by-caller note: * Retryable because {@link #checkLock(Connection, long)} is */ @Override @RetrySemantics.SafeToRetry public LockResponse checkLock(CheckLockRequest rqst) throws NoSuchTxnException, NoSuchLockException, TxnAbortedException, MetaException { try { Connection dbConn = null; long extLockId = rqst.getLockid(); try { lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); // Heartbeat on the lockid first, to assure that our lock is still valid. // Then look up the lock info (hopefully in the cache). If these locks // are associated with a transaction then heartbeat on that as well. LockInfo info = getTxnIdFromLockId(dbConn, extLockId); if (info == null) { throw new NoSuchLockException("No such lock " + JavaUtils.lockIdToString(extLockId)); } if (info.txnId > 0) { heartbeatTxn(dbConn, info.txnId); } else { heartbeatLock(dbConn, extLockId); } //todo: strictly speaking there is a bug here. heartbeat*() commits but both heartbeat and //checkLock() are in the same retry block, so if checkLock() throws, heartbeat is also retired //extra heartbeat is logically harmless, but ... return checkLock(dbConn, extLockId); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "checkLock(" + rqst + " )"); throw new MetaException("Unable to update transaction database " + JavaUtils.lockIdToString(extLockId) + " " + StringUtils.stringifyException(e)); } finally { closeDbConn(dbConn); unlockInternal(); } } catch (RetryException e) { return checkLock(rqst); } } /** * This would have been made simpler if all locks were associated with a txn. Then only txn needs to * be heartbeated, committed, etc. no need for client to track individual locks. * When removing locks not associated with txn this potentially conflicts with * heartbeat/performTimeout which are update/delete of HIVE_LOCKS thus will be locked as needed by db. * since this only removes from HIVE_LOCKS at worst some lock acquire is delayed */ @RetrySemantics.Idempotent public void unlock(UnlockRequest rqst) throws NoSuchLockException, TxnOpenException, MetaException { try { Connection dbConn = null; Statement stmt = null; long extLockId = rqst.getLockid(); try { /** * This method is logically like commit for read-only auto commit queries. * READ_COMMITTED since this only has 1 delete statement and no new entries with the * same hl_lock_ext_id can be added, i.e. all rows with a given hl_lock_ext_id are * created in a single atomic operation. * Theoretically, this competes with {@link #lock(org.apache.hadoop.hive.metastore.api.LockRequest)} * but hl_lock_ext_id is not known until that method returns. * Also competes with {@link #checkLock(org.apache.hadoop.hive.metastore.api.CheckLockRequest)} * but using SERIALIZABLE doesn't materially change the interaction. * If "delete" stmt misses, additional logic is best effort to produce meaningful error msg. */ dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); //hl_txnid <> 0 means it's associated with a transaction String s = "delete from HIVE_LOCKS where hl_lock_ext_id = " + extLockId + " AND (hl_txnid = 0 OR" + " (hl_txnid <> 0 AND hl_lock_state = '" + LOCK_WAITING + "'))"; //(hl_txnid <> 0 AND hl_lock_state = '" + LOCK_WAITING + "') is for multi-statement txns where //some query attempted to lock (thus LOCK_WAITING state) but is giving up due to timeout for example LOG.debug("Going to execute update <" + s + ">"); int rc = stmt.executeUpdate(s); if (rc < 1) { LOG.debug("Going to rollback"); dbConn.rollback(); LockInfo info = getTxnIdFromLockId(dbConn, extLockId); if (info == null) { //didn't find any lock with extLockId but at ReadCommitted there is a possibility that //it existed when above delete ran but it didn't have the expected state. LOG.info("No lock in " + LOCK_WAITING + " mode found for unlock(" + JavaUtils.lockIdToString(rqst.getLockid()) + ")"); //bail here to make the operation idempotent return; } if (info.txnId != 0) { String msg = "Unlocking locks associated with transaction not permitted. " + info; //if a lock is associated with a txn we can only "unlock" if if it's in WAITING state // which really means that the caller wants to give up waiting for the lock LOG.error(msg); throw new TxnOpenException(msg); } if (info.txnId == 0) { //we didn't see this lock when running DELETE stmt above but now it showed up //so should "should never happen" happened... String msg = "Found lock in unexpected state " + info; LOG.error(msg); throw new MetaException(msg); } } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "unlock(" + rqst + ")"); throw new MetaException("Unable to update transaction database " + JavaUtils.lockIdToString(extLockId) + " " + StringUtils.stringifyException(e)); } finally { closeStmt(stmt); closeDbConn(dbConn); } } catch (RetryException e) { unlock(rqst); } } /** * used to sort entries in {@link org.apache.hadoop.hive.metastore.api.ShowLocksResponse} */ private static class LockInfoExt extends LockInfo { private final ShowLocksResponseElement e; LockInfoExt(ShowLocksResponseElement e) { super(e); this.e = e; } } @RetrySemantics.ReadOnly public ShowLocksResponse showLocks(ShowLocksRequest rqst) throws MetaException { try { Connection dbConn = null; ShowLocksResponse rsp = new ShowLocksResponse(); List<ShowLocksResponseElement> elems = new ArrayList<ShowLocksResponseElement>(); List<LockInfoExt> sortedList = new ArrayList<LockInfoExt>(); Statement stmt = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select hl_lock_ext_id, hl_txnid, hl_db, hl_table, hl_partition, hl_lock_state, " + "hl_lock_type, hl_last_heartbeat, hl_acquired_at, hl_user, hl_host, hl_lock_int_id," + "hl_blockedby_ext_id, hl_blockedby_int_id, hl_agent_info from HIVE_LOCKS"; // Some filters may have been specified in the SHOW LOCKS statement. Add them to the query. String dbName = rqst.getDbname(); String tableName = rqst.getTablename(); String partName = rqst.getPartname(); StringBuilder filter = new StringBuilder(); if (dbName != null && !dbName.isEmpty()) { filter.append("hl_db=").append(quoteString(dbName)); } if (tableName != null && !tableName.isEmpty()) { if (filter.length() > 0) { filter.append(" and "); } filter.append("hl_table=").append(quoteString(tableName)); } if (partName != null && !partName.isEmpty()) { if (filter.length() > 0) { filter.append(" and "); } filter.append("hl_partition=").append(quoteString(partName)); } String whereClause = filter.toString(); if (!whereClause.isEmpty()) { s = s + " where " + whereClause; } LOG.debug("Doing to execute query <" + s + ">"); ResultSet rs = stmt.executeQuery(s); while (rs.next()) { ShowLocksResponseElement e = new ShowLocksResponseElement(); e.setLockid(rs.getLong(1)); long txnid = rs.getLong(2); if (!rs.wasNull()) e.setTxnid(txnid); e.setDbname(rs.getString(3)); e.setTablename(rs.getString(4)); String partition = rs.getString(5); if (partition != null) e.setPartname(partition); switch (rs.getString(6).charAt(0)) { case LOCK_ACQUIRED: e.setState(LockState.ACQUIRED); break; case LOCK_WAITING: e.setState(LockState.WAITING); break; default: throw new MetaException("Unknown lock state " + rs.getString(6).charAt(0)); } switch (rs.getString(7).charAt(0)) { case LOCK_SEMI_SHARED: e.setType(LockType.SHARED_WRITE); break; case LOCK_EXCLUSIVE: e.setType(LockType.EXCLUSIVE); break; case LOCK_SHARED: e.setType(LockType.SHARED_READ); break; default: throw new MetaException("Unknown lock type " + rs.getString(6).charAt(0)); } e.setLastheartbeat(rs.getLong(8)); long acquiredAt = rs.getLong(9); if (!rs.wasNull()) e.setAcquiredat(acquiredAt); e.setUser(rs.getString(10)); e.setHostname(rs.getString(11)); e.setLockIdInternal(rs.getLong(12)); long id = rs.getLong(13); if (!rs.wasNull()) { e.setBlockedByExtId(id); } id = rs.getLong(14); if (!rs.wasNull()) { e.setBlockedByIntId(id); } e.setAgentInfo(rs.getString(15)); sortedList.add(new LockInfoExt(e)); } LOG.debug("Going to rollback"); dbConn.rollback(); } catch (SQLException e) { checkRetryable(dbConn, e, "showLocks(" + rqst + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { closeStmt(stmt); closeDbConn(dbConn); } //this ensures that "SHOW LOCKS" prints the locks in the same order as they are examined //by checkLock() - makes diagnostics easier. Collections.sort(sortedList, new LockInfoComparator()); for (LockInfoExt lockInfoExt : sortedList) { elems.add(lockInfoExt.e); } rsp.setLocks(elems); return rsp; } catch (RetryException e) { return showLocks(rqst); } } /** * {@code ids} should only have txnid or lockid but not both, ideally. * Currently DBTxnManager.heartbeat() enforces this. */ @Override @RetrySemantics.SafeToRetry public void heartbeat(HeartbeatRequest ids) throws NoSuchTxnException, NoSuchLockException, TxnAbortedException, MetaException { try { Connection dbConn = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); heartbeatLock(dbConn, ids.getLockid()); heartbeatTxn(dbConn, ids.getTxnid()); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "heartbeat(" + ids + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { closeDbConn(dbConn); } } catch (RetryException e) { heartbeat(ids); } } @Override @RetrySemantics.SafeToRetry public HeartbeatTxnRangeResponse heartbeatTxnRange(HeartbeatTxnRangeRequest rqst) throws MetaException { try { Connection dbConn = null; Statement stmt = null; HeartbeatTxnRangeResponse rsp = new HeartbeatTxnRangeResponse(); Set<Long> nosuch = new HashSet<Long>(); Set<Long> aborted = new HashSet<Long>(); rsp.setNosuch(nosuch); rsp.setAborted(aborted); try { /** * READ_COMMITTED is sufficient since {@link #heartbeatTxn(java.sql.Connection, long)} * only has 1 update statement in it and * we only update existing txns, i.e. nothing can add additional txns that this operation * would care about (which would have required SERIALIZABLE) */ dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); /*do fast path first (in 1 statement) if doesn't work, rollback and do the long version*/ stmt = dbConn.createStatement(); List<String> queries = new ArrayList<>(); int numTxnsToHeartbeat = (int) (rqst.getMax() - rqst.getMin() + 1); List<Long> txnIds = new ArrayList<>(numTxnsToHeartbeat); for (long txn = rqst.getMin(); txn <= rqst.getMax(); txn++) { txnIds.add(txn); } TxnUtils.buildQueryWithINClause(conf, queries, new StringBuilder("update TXNS set txn_last_heartbeat = " + getDbTime(dbConn) + " where txn_state = " + quoteChar(TXN_OPEN) + " and "), new StringBuilder(""), txnIds, "txn_id", true, false); int updateCnt = 0; for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); updateCnt += stmt.executeUpdate(query); } if (updateCnt == numTxnsToHeartbeat) { //fast pass worked, i.e. all txns we were asked to heartbeat were Open as expected dbConn.commit(); return rsp; } //if here, do the slow path so that we can return info txns which were not in expected state dbConn.rollback(); for (long txn = rqst.getMin(); txn <= rqst.getMax(); txn++) { try { heartbeatTxn(dbConn, txn); } catch (NoSuchTxnException e) { nosuch.add(txn); } catch (TxnAbortedException e) { aborted.add(txn); } } return rsp; } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "heartbeatTxnRange(" + rqst + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { close(null, stmt, dbConn); } } catch (RetryException e) { return heartbeatTxnRange(rqst); } } long generateCompactionQueueId(Statement stmt) throws SQLException, MetaException { // Get the id for the next entry in the queue String s = sqlGenerator.addForUpdateClause("select ncq_next from NEXT_COMPACTION_QUEUE_ID"); LOG.debug("going to execute query <" + s + ">"); try (ResultSet rs = stmt.executeQuery(s)) { if (!rs.next()) { throw new IllegalStateException("Transaction tables not properly initiated, " + "no record found in next_compaction_queue_id"); } long id = rs.getLong(1); s = "update NEXT_COMPACTION_QUEUE_ID set ncq_next = " + (id + 1); LOG.debug("Going to execute update <" + s + ">"); stmt.executeUpdate(s); return id; } } @Override @RetrySemantics.Idempotent public CompactionResponse compact(CompactionRequest rqst) throws MetaException { // Put a compaction request in the queue. try { Connection dbConn = null; Statement stmt = null; TxnStore.MutexAPI.LockHandle handle = null; try { lockInternal(); /** * MUTEX_KEY.CompactionScheduler lock ensures that there is only 1 entry in * Initiated/Working state for any resource. This ensures that we don't run concurrent * compactions for any resource. */ handle = getMutexAPI().acquireLock(MUTEX_KEY.CompactionScheduler.name()); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); long id = generateCompactionQueueId(stmt); StringBuilder sb = new StringBuilder("select cq_id, cq_state from COMPACTION_QUEUE where") .append(" cq_state IN(").append(quoteChar(INITIATED_STATE)).append(",") .append(quoteChar(WORKING_STATE)).append(") AND cq_database=") .append(quoteString(rqst.getDbname())).append(" AND cq_table=") .append(quoteString(rqst.getTablename())).append(" AND "); if (rqst.getPartitionname() == null) { sb.append("cq_partition is null"); } else { sb.append("cq_partition=").append(quoteString(rqst.getPartitionname())); } LOG.debug("Going to execute query <" + sb.toString() + ">"); ResultSet rs = stmt.executeQuery(sb.toString()); if (rs.next()) { long enqueuedId = rs.getLong(1); String state = compactorStateToResponse(rs.getString(2).charAt(0)); LOG.info("Ignoring request to compact " + rqst.getDbname() + "/" + rqst.getTablename() + "/" + rqst.getPartitionname() + " since it is already " + quoteString(state) + " with id=" + enqueuedId); return new CompactionResponse(enqueuedId, state, false); } close(rs); StringBuilder buf = new StringBuilder( "insert into COMPACTION_QUEUE (cq_id, cq_database, " + "cq_table, "); String partName = rqst.getPartitionname(); if (partName != null) buf.append("cq_partition, "); buf.append("cq_state, cq_type"); if (rqst.getProperties() != null) { buf.append(", cq_tblproperties"); } if (rqst.getRunas() != null) buf.append(", cq_run_as"); buf.append(") values ("); buf.append(id); buf.append(", '"); buf.append(rqst.getDbname()); buf.append("', '"); buf.append(rqst.getTablename()); buf.append("', '"); if (partName != null) { buf.append(partName); buf.append("', '"); } buf.append(INITIATED_STATE); buf.append("', '"); switch (rqst.getType()) { case MAJOR: buf.append(MAJOR_TYPE); break; case MINOR: buf.append(MINOR_TYPE); break; default: LOG.debug("Going to rollback"); dbConn.rollback(); throw new MetaException("Unexpected compaction type " + rqst.getType().toString()); } if (rqst.getProperties() != null) { buf.append("', '"); buf.append(new StringableMap(rqst.getProperties()).toString()); } if (rqst.getRunas() != null) { buf.append("', '"); buf.append(rqst.getRunas()); } buf.append("')"); String s = buf.toString(); LOG.debug("Going to execute update <" + s + ">"); stmt.executeUpdate(s); LOG.debug("Going to commit"); dbConn.commit(); return new CompactionResponse(id, INITIATED_RESPONSE, true); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "compact(" + rqst + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { closeStmt(stmt); closeDbConn(dbConn); if (handle != null) { handle.releaseLocks(); } unlockInternal(); } } catch (RetryException e) { return compact(rqst); } } private static String compactorStateToResponse(char s) { switch (s) { case INITIATED_STATE: return INITIATED_RESPONSE; case WORKING_STATE: return WORKING_RESPONSE; case READY_FOR_CLEANING: return CLEANING_RESPONSE; case FAILED_STATE: return FAILED_RESPONSE; case SUCCEEDED_STATE: return SUCCEEDED_RESPONSE; case ATTEMPTED_STATE: return ATTEMPTED_RESPONSE; default: return Character.toString(s); } } @RetrySemantics.ReadOnly public ShowCompactResponse showCompact(ShowCompactRequest rqst) throws MetaException { ShowCompactResponse response = new ShowCompactResponse(new ArrayList<ShowCompactResponseElement>()); Connection dbConn = null; Statement stmt = null; try { try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select cq_database, cq_table, cq_partition, cq_state, cq_type, cq_worker_id, " + //-1 because 'null' literal doesn't work for all DBs... "cq_start, -1 cc_end, cq_run_as, cq_hadoop_job_id, cq_id from COMPACTION_QUEUE union all " + "select cc_database, cc_table, cc_partition, cc_state, cc_type, cc_worker_id, " + "cc_start, cc_end, cc_run_as, cc_hadoop_job_id, cc_id from COMPLETED_COMPACTIONS"; //what I want is order by cc_end desc, cc_start asc (but derby has a bug https://issues.apache.org/jira/browse/DERBY-6013) //to sort so that currently running jobs are at the end of the list (bottom of screen) //and currently running ones are in sorted by start time //w/o order by likely currently running compactions will be first (LHS of Union) LOG.debug("Going to execute query <" + s + ">"); ResultSet rs = stmt.executeQuery(s); while (rs.next()) { ShowCompactResponseElement e = new ShowCompactResponseElement(); e.setDbname(rs.getString(1)); e.setTablename(rs.getString(2)); e.setPartitionname(rs.getString(3)); e.setState(compactorStateToResponse(rs.getString(4).charAt(0))); switch (rs.getString(5).charAt(0)) { case MAJOR_TYPE: e.setType(CompactionType.MAJOR); break; case MINOR_TYPE: e.setType(CompactionType.MINOR); break; default: //do nothing to handle RU/D if we add another status } e.setWorkerid(rs.getString(6)); long start = rs.getLong(7); if (!rs.wasNull()) { e.setStart(start); } long endTime = rs.getLong(8); if (endTime != -1) { e.setEndTime(endTime); } e.setRunAs(rs.getString(9)); e.setHadoopJobId(rs.getString(10)); e.setId(rs.getLong(11)); response.addToCompacts(e); } LOG.debug("Going to rollback"); dbConn.rollback(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "showCompact(" + rqst + ")"); throw new MetaException( "Unable to select from transaction database " + StringUtils.stringifyException(e)); } finally { closeStmt(stmt); closeDbConn(dbConn); } return response; } catch (RetryException e) { return showCompact(rqst); } } private static void shouldNeverHappen(long txnid) { throw new RuntimeException("This should never happen: " + JavaUtils.txnIdToString(txnid)); } private static void shouldNeverHappen(long txnid, long extLockId, long intLockId) { throw new RuntimeException("This should never happen: " + JavaUtils.txnIdToString(txnid) + " " + JavaUtils.lockIdToString(extLockId) + " " + intLockId); } /** * Retry-by-caller note: * This may be retried after dbConn.commit. At worst, it will create duplicate entries in * TXN_COMPONENTS which won't affect anything. See more comments in {@link #commitTxn(CommitTxnRequest)} */ @Override @RetrySemantics.SafeToRetry public void addDynamicPartitions(AddDynamicPartitions rqst) throws NoSuchTxnException, TxnAbortedException, MetaException { Connection dbConn = null; Statement stmt = null; ResultSet lockHandle = null; ResultSet rs = null; try { try { lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); lockHandle = lockTransactionRecord(stmt, rqst.getTxnid(), TXN_OPEN); if (lockHandle == null) { //ensures txn is still there and in expected state ensureValidTxn(dbConn, rqst.getTxnid(), stmt); shouldNeverHappen(rqst.getTxnid()); } //for RU this may be null so we should default it to 'u' which is most restrictive OpertaionType ot = OpertaionType.UPDATE; if (rqst.isSetOperationType()) { ot = OpertaionType.fromDataOperationType(rqst.getOperationType()); } List<String> rows = new ArrayList<>(); for (String partName : rqst.getPartitionnames()) { rows.add(rqst.getTxnid() + "," + quoteString(rqst.getDbname()) + "," + quoteString(rqst.getTablename()) + "," + quoteString(partName) + "," + quoteChar(ot.sqlConst)); } int modCount = 0; //record partitions that were written to List<String> queries = sqlGenerator.createInsertValuesStmt( "TXN_COMPONENTS (tc_txnid, tc_database, tc_table, tc_partition, tc_operation_type)", rows); for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); modCount = stmt.executeUpdate(query); } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "addDynamicPartitions(" + rqst + ")"); throw new MetaException( "Unable to insert into from transaction database " + StringUtils.stringifyException(e)); } finally { close(lockHandle, stmt, dbConn); unlockInternal(); } } catch (RetryException e) { addDynamicPartitions(rqst); } } /** * Clean up corresponding records in metastore tables when corresponding object is dropped, * specifically: TXN_COMPONENTS, COMPLETED_TXN_COMPONENTS, COMPACTION_QUEUE, COMPLETED_COMPACTIONS * Retry-by-caller note: this is only idempotent assuming it's only called by dropTable/Db/etc * operations. */ @Override @RetrySemantics.Idempotent public void cleanupRecords(HiveObjectType type, Database db, Table table, Iterator<Partition> partitionIterator) throws MetaException { try { Connection dbConn = null; Statement stmt = null; try { String dbName; String tblName; dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); List<String> queries = new ArrayList<String>(); StringBuilder buff = new StringBuilder(); switch (type) { case DATABASE: dbName = db.getName(); buff.append("delete from TXN_COMPONENTS where tc_database='"); buff.append(dbName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_TXN_COMPONENTS where ctc_database='"); buff.append(dbName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPACTION_QUEUE where cq_database='"); buff.append(dbName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_COMPACTIONS where cc_database='"); buff.append(dbName); buff.append("'"); queries.add(buff.toString()); break; case TABLE: dbName = table.getDbName(); tblName = table.getTableName(); buff.append("delete from TXN_COMPONENTS where tc_database='"); buff.append(dbName); buff.append("' and tc_table='"); buff.append(tblName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_TXN_COMPONENTS where ctc_database='"); buff.append(dbName); buff.append("' and ctc_table='"); buff.append(tblName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPACTION_QUEUE where cq_database='"); buff.append(dbName); buff.append("' and cq_table='"); buff.append(tblName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_COMPACTIONS where cc_database='"); buff.append(dbName); buff.append("' and cc_table='"); buff.append(tblName); buff.append("'"); queries.add(buff.toString()); break; case PARTITION: dbName = table.getDbName(); tblName = table.getTableName(); List<FieldSchema> partCols = table.getPartitionKeys(); // partition columns List<String> partVals; // partition values String partName; while (partitionIterator.hasNext()) { Partition p = partitionIterator.next(); partVals = p.getValues(); partName = Warehouse.makePartName(partCols, partVals); buff.append("delete from TXN_COMPONENTS where tc_database='"); buff.append(dbName); buff.append("' and tc_table='"); buff.append(tblName); buff.append("' and tc_partition='"); buff.append(partName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_TXN_COMPONENTS where ctc_database='"); buff.append(dbName); buff.append("' and ctc_table='"); buff.append(tblName); buff.append("' and ctc_partition='"); buff.append(partName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPACTION_QUEUE where cq_database='"); buff.append(dbName); buff.append("' and cq_table='"); buff.append(tblName); buff.append("' and cq_partition='"); buff.append(partName); buff.append("'"); queries.add(buff.toString()); buff.setLength(0); buff.append("delete from COMPLETED_COMPACTIONS where cc_database='"); buff.append(dbName); buff.append("' and cc_table='"); buff.append(tblName); buff.append("' and cc_partition='"); buff.append(partName); buff.append("'"); queries.add(buff.toString()); } break; default: throw new MetaException("Invalid object type for cleanup: " + type); } for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); stmt.executeUpdate(query); } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); checkRetryable(dbConn, e, "cleanupRecords"); if (e.getMessage().contains("does not exist")) { LOG.warn("Cannot perform cleanup since metastore table does not exist"); } else { throw new MetaException("Unable to clean up " + StringUtils.stringifyException(e)); } } finally { closeStmt(stmt); closeDbConn(dbConn); } } catch (RetryException e) { cleanupRecords(type, db, table, partitionIterator); } } /** * For testing only, do not use. */ @VisibleForTesting public int numLocksInLockTable() throws SQLException, MetaException { Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select count(*) from HIVE_LOCKS"; LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); rs.next(); int rc = rs.getInt(1); // Necessary to clean up the transaction in the db. dbConn.rollback(); return rc; } finally { close(rs, stmt, dbConn); } } /** * For testing only, do not use. */ public long setTimeout(long milliseconds) { long previous_timeout = timeout; timeout = milliseconds; return previous_timeout; } protected class RetryException extends Exception { } Connection getDbConn(int isolationLevel) throws SQLException { return getDbConn(isolationLevel, connPool); } private Connection getDbConn(int isolationLevel, DataSource connPool) throws SQLException { int rc = doRetryOnConnPool ? 10 : 1; Connection dbConn = null; while (true) { try { dbConn = connPool.getConnection(); dbConn.setAutoCommit(false); dbConn.setTransactionIsolation(isolationLevel); return dbConn; } catch (SQLException e) { closeDbConn(dbConn); if ((--rc) <= 0) throw e; LOG.error("There is a problem with a connection from the pool, retrying(rc=" + rc + "): " + getMessage(e), e); } } } static void rollbackDBConn(Connection dbConn) { try { if (dbConn != null && !dbConn.isClosed()) dbConn.rollback(); } catch (SQLException e) { LOG.warn("Failed to rollback db connection " + getMessage(e)); } } protected static void closeDbConn(Connection dbConn) { try { if (dbConn != null && !dbConn.isClosed()) { dbConn.close(); } } catch (SQLException e) { LOG.warn("Failed to close db connection " + getMessage(e)); } } /** * Close statement instance. * @param stmt statement instance. */ protected static void closeStmt(Statement stmt) { try { if (stmt != null && !stmt.isClosed()) stmt.close(); } catch (SQLException e) { LOG.warn("Failed to close statement " + getMessage(e)); } } /** * Close the ResultSet. * @param rs may be {@code null} */ static void close(ResultSet rs) { try { if (rs != null && !rs.isClosed()) { rs.close(); } } catch (SQLException ex) { LOG.warn("Failed to close statement " + getMessage(ex)); } } /** * Close all 3 JDBC artifacts in order: {@code rs stmt dbConn} */ static void close(ResultSet rs, Statement stmt, Connection dbConn) { close(rs); closeStmt(stmt); closeDbConn(dbConn); } /** * Determine if an exception was such that it makes sense to retry. Unfortunately there is no standard way to do * this, so we have to inspect the error messages and catch the telltale signs for each * different database. This method will throw {@code RetryException} * if the error is retry-able. * @param conn database connection * @param e exception that was thrown. * @param caller name of the method calling this (and other info useful to log) * @throws org.apache.hadoop.hive.metastore.txn.TxnHandler.RetryException when the operation should be retried */ protected void checkRetryable(Connection conn, SQLException e, String caller) throws RetryException, MetaException { // If you change this function, remove the @Ignore from TestTxnHandler.deadlockIsDetected() // to test these changes. // MySQL and MSSQL use 40001 as the state code for rollback. Postgres uses 40001 and 40P01. // Oracle seems to return different SQLStates and messages each time, // so I've tried to capture the different error messages (there appear to be fewer different // error messages than SQL states). // Derby and newer MySQL driver use the new SQLTransactionRollbackException boolean sendRetrySignal = false; try { if (dbProduct == null) { throw new IllegalStateException("DB Type not determined yet."); } if (DatabaseProduct.isDeadlock(dbProduct, e)) { if (deadlockCnt++ < ALLOWED_REPEATED_DEADLOCKS) { long waitInterval = deadlockRetryInterval * deadlockCnt; LOG.warn("Deadlock detected in " + caller + ". Will wait " + waitInterval + "ms try again up to " + (ALLOWED_REPEATED_DEADLOCKS - deadlockCnt + 1) + " times."); // Pause for a just a bit for retrying to avoid immediately jumping back into the deadlock. try { Thread.sleep(waitInterval); } catch (InterruptedException ie) { // NOP } sendRetrySignal = true; } else { LOG.error("Too many repeated deadlocks in " + caller + ", giving up."); } } else if (isRetryable(conf, e)) { //in MSSQL this means Communication Link Failure if (retryNum++ < retryLimit) { LOG.warn("Retryable error detected in " + caller + ". Will wait " + retryInterval + "ms and retry up to " + (retryLimit - retryNum + 1) + " times. Error: " + getMessage(e)); try { Thread.sleep(retryInterval); } catch (InterruptedException ex) { // } sendRetrySignal = true; } else { LOG.error("Fatal error in " + caller + ". Retry limit (" + retryLimit + ") reached. Last error: " + getMessage(e)); } } else { //make sure we know we saw an error that we don't recognize LOG.info("Non-retryable error in " + caller + " : " + getMessage(e)); } } finally { /*if this method ends with anything except a retry signal, the caller should fail the operation and propagate the error up to the its caller (Metastore client); thus must reset retry counters*/ if (!sendRetrySignal) { deadlockCnt = 0; retryNum = 0; } } if (sendRetrySignal) { throw new RetryException(); } } /** * Determine the current time, using the RDBMS as a source of truth * @param conn database connection * @return current time in milliseconds * @throws org.apache.hadoop.hive.metastore.api.MetaException if the time cannot be determined */ protected long getDbTime(Connection conn) throws MetaException { Statement stmt = null; try { stmt = conn.createStatement(); String s; switch (dbProduct) { case DERBY: s = "values current_timestamp"; break; case MYSQL: case POSTGRES: case SQLSERVER: s = "select current_timestamp"; break; case ORACLE: s = "select current_timestamp from dual"; break; default: String msg = "Unknown database product: " + dbProduct.toString(); LOG.error(msg); throw new MetaException(msg); } LOG.debug("Going to execute query <" + s + ">"); ResultSet rs = stmt.executeQuery(s); if (!rs.next()) throw new MetaException("No results from date query"); return rs.getTimestamp(1).getTime(); } catch (SQLException e) { String msg = "Unable to determine current time: " + e.getMessage(); LOG.error(msg); throw new MetaException(msg); } finally { closeStmt(stmt); } } /** * Determine the String that should be used to quote identifiers. * @param conn Active connection * @return quotes * @throws SQLException */ protected String getIdentifierQuoteString(Connection conn) throws SQLException { if (identifierQuoteString == null) { identifierQuoteString = conn.getMetaData().getIdentifierQuoteString(); } return identifierQuoteString; } private void determineDatabaseProduct(Connection conn) { if (dbProduct != null) return; try { String s = conn.getMetaData().getDatabaseProductName(); dbProduct = DatabaseProduct.determineDatabaseProduct(s); if (dbProduct == DatabaseProduct.OTHER) { String msg = "Unrecognized database product name <" + s + ">"; LOG.error(msg); throw new IllegalStateException(msg); } } catch (SQLException e) { String msg = "Unable to get database product name"; LOG.error(msg, e); throw new IllegalStateException(msg, e); } } private static class LockInfo { private final long extLockId; private final long intLockId; //0 means there is no transaction, i.e. it a select statement which is not part of //explicit transaction or a IUD statement that is not writing to ACID table private final long txnId; private final String db; private final String table; private final String partition; private final LockState state; private final LockType type; // Assumes the result set is set to a valid row LockInfo(ResultSet rs) throws SQLException, MetaException { extLockId = rs.getLong("hl_lock_ext_id"); // can't be null intLockId = rs.getLong("hl_lock_int_id"); // can't be null db = rs.getString("hl_db"); // can't be null String t = rs.getString("hl_table"); table = (rs.wasNull() ? null : t); String p = rs.getString("hl_partition"); partition = (rs.wasNull() ? null : p); switch (rs.getString("hl_lock_state").charAt(0)) { case LOCK_WAITING: state = LockState.WAITING; break; case LOCK_ACQUIRED: state = LockState.ACQUIRED; break; default: throw new MetaException("Unknown lock state " + rs.getString("hl_lock_state").charAt(0)); } switch (rs.getString("hl_lock_type").charAt(0)) { case LOCK_EXCLUSIVE: type = LockType.EXCLUSIVE; break; case LOCK_SHARED: type = LockType.SHARED_READ; break; case LOCK_SEMI_SHARED: type = LockType.SHARED_WRITE; break; default: throw new MetaException("Unknown lock type " + rs.getString("hl_lock_type").charAt(0)); } txnId = rs.getLong("hl_txnid");//returns 0 if value is NULL } LockInfo(ShowLocksResponseElement e) { extLockId = e.getLockid(); intLockId = e.getLockIdInternal(); txnId = e.getTxnid(); db = e.getDbname(); table = e.getTablename(); partition = e.getPartname(); state = e.getState(); type = e.getType(); } public boolean equals(Object other) { if (!(other instanceof LockInfo)) return false; LockInfo o = (LockInfo) other; // Lock ids are unique across the system. return extLockId == o.extLockId && intLockId == o.intLockId; } @Override public String toString() { return JavaUtils.lockIdToString(extLockId) + " intLockId:" + intLockId + " " + JavaUtils.txnIdToString(txnId) + " db:" + db + " table:" + table + " partition:" + partition + " state:" + (state == null ? "null" : state.toString()) + " type:" + (type == null ? "null" : type.toString()); } private boolean isDbLock() { return db != null && table == null && partition == null; } private boolean isTableLock() { return db != null && table != null && partition == null; } private boolean isPartitionLock() { return !(isDbLock() || isTableLock()); } } private static class LockInfoComparator implements Comparator<LockInfo> { private static final LockTypeComparator lockTypeComparator = new LockTypeComparator(); public boolean equals(Object other) { return this == other; } public int compare(LockInfo info1, LockInfo info2) { // We sort by state (acquired vs waiting) and then by LockType, then by id if (info1.state == LockState.ACQUIRED && info2.state != LockState.ACQUIRED) { return -1; } if (info1.state != LockState.ACQUIRED && info2.state == LockState.ACQUIRED) { return 1; } int sortByType = lockTypeComparator.compare(info1.type, info2.type); if (sortByType != 0) { return sortByType; } if (info1.extLockId < info2.extLockId) { return -1; } else if (info1.extLockId > info2.extLockId) { return 1; } else { if (info1.intLockId < info2.intLockId) { return -1; } else if (info1.intLockId > info2.intLockId) { return 1; } else { return 0; } } } } /** * Sort more restrictive locks after less restrictive ones */ private final static class LockTypeComparator implements Comparator<LockType> { public boolean equals(Object other) { return this == other; } public int compare(LockType t1, LockType t2) { switch (t1) { case EXCLUSIVE: if (t2 == LockType.EXCLUSIVE) { return 0; } return 1; case SHARED_WRITE: switch (t2) { case EXCLUSIVE: return -1; case SHARED_WRITE: return 0; case SHARED_READ: return 1; default: throw new RuntimeException("Unexpected LockType: " + t2); } case SHARED_READ: if (t2 == LockType.SHARED_READ) { return 0; } return -1; default: throw new RuntimeException("Unexpected LockType: " + t1); } } } private enum LockAction { ACQUIRE, WAIT, KEEP_LOOKING } // A jump table to figure out whether to wait, acquire, // or keep looking . Since // java doesn't have function pointers (grumble grumble) we store a // character that we'll use to determine which function to call. // The table maps the lock type of the lock we are looking to acquire to // the lock type of the lock we are checking to the lock state of the lock // we are checking to the desired action. private static Map<LockType, Map<LockType, Map<LockState, LockAction>>> jumpTable; private void checkQFileTestHack() { boolean hackOn = HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_IN_TEST) || HiveConf.getBoolVar(conf, HiveConf.ConfVars.HIVE_IN_TEZ_TEST); if (hackOn) { LOG.info("Hacking in canned values for transaction manager"); // Set up the transaction/locking db in the derby metastore TxnDbUtil.setConfValues(conf); try { TxnDbUtil.prepDb(); } catch (Exception e) { // We may have already created the tables and thus don't need to redo it. if (e.getMessage() != null && !e.getMessage().contains("already exists")) { throw new RuntimeException( "Unable to set up transaction database for" + " testing: " + e.getMessage(), e); } } } } private int abortTxns(Connection dbConn, List<Long> txnids, boolean isStrict) throws SQLException { return abortTxns(dbConn, txnids, -1, isStrict); } /** * TODO: expose this as an operation to client. Useful for streaming API to abort all remaining * trasnactions in a batch on IOExceptions. * Caller must rollback the transaction if not all transactions were aborted since this will not * attempt to delete associated locks in this case. * * @param dbConn An active connection * @param txnids list of transactions to abort * @param max_heartbeat value used by {@link #performTimeOuts()} to ensure this doesn't Abort txn which were * hearbetated after #performTimeOuts() select and this operation. * @param isStrict true for strict mode, false for best-effort mode. * In strict mode, if all txns are not successfully aborted, then the count of * updated ones will be returned and the caller will roll back. * In best-effort mode, we will ignore that fact and continue deleting the locks. * @return Number of aborted transactions * @throws SQLException */ private int abortTxns(Connection dbConn, List<Long> txnids, long max_heartbeat, boolean isStrict) throws SQLException { Statement stmt = null; int updateCnt = 0; if (txnids.isEmpty()) { return 0; } try { stmt = dbConn.createStatement(); //This is an update statement, thus at any Isolation level will take Write locks so will block //all other ops using S4U on TXNS row. List<String> queries = new ArrayList<String>(); StringBuilder prefix = new StringBuilder(); StringBuilder suffix = new StringBuilder(); prefix.append("update TXNS set txn_state = " + quoteChar(TXN_ABORTED) + " where txn_state = " + quoteChar(TXN_OPEN) + " and "); if (max_heartbeat > 0) { suffix.append(" and txn_last_heartbeat < ").append(max_heartbeat); } else { suffix.append(""); } TxnUtils.buildQueryWithINClause(conf, queries, prefix, suffix, txnids, "txn_id", true, false); for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); updateCnt += stmt.executeUpdate(query); } if (updateCnt < txnids.size() && isStrict) { /** * have to bail in this case since we don't know which transactions were not Aborted and * thus don't know which locks to delete * This may happen due to a race between {@link #heartbeat(HeartbeatRequest)} operation and * {@link #performTimeOuts()} */ return updateCnt; } queries.clear(); prefix.setLength(0); suffix.setLength(0); prefix.append("delete from HIVE_LOCKS where "); suffix.append(""); TxnUtils.buildQueryWithINClause(conf, queries, prefix, suffix, txnids, "hl_txnid", false, false); for (String query : queries) { LOG.debug("Going to execute update <" + query + ">"); int rc = stmt.executeUpdate(query); LOG.debug("Removed " + rc + " records from HIVE_LOCKS"); } } finally { closeStmt(stmt); } return updateCnt; } private static boolean isValidTxn(long txnId) { return txnId != 0; } /** * Lock acquisition is meant to be fair, so every lock can only block on some lock with smaller * hl_lock_ext_id by only checking earlier locks. * * For any given SQL statment all locks required by it are grouped under single extLockId and are * granted all at once or all locks wait. * * This is expected to run at READ_COMMITTED. * * Note: this calls acquire() for (extLockId,intLockId) but extLockId is the same and we either take * all locks for given extLockId or none. Would be more efficient to update state on all locks * at once. Semantics are the same since this is all part of the same txn. * * If there is a concurrent commitTxn/rollbackTxn, those can only remove rows from HIVE_LOCKS. * If they happen to be for the same txnid, there will be a WW conflict (in MS DB), if different txnid, * checkLock() will in the worst case keep locks in Waiting state a little longer. */ @RetrySemantics.SafeToRetry("See @SafeToRetry") private LockResponse checkLock(Connection dbConn, long extLockId) throws NoSuchLockException, NoSuchTxnException, TxnAbortedException, MetaException, SQLException { TxnStore.MutexAPI.LockHandle handle = null; Statement stmt = null; ResultSet rs = null; LockResponse response = new LockResponse(); /** * todo: Longer term we should pass this from client somehow - this would be an optimization; once * that is in place make sure to build and test "writeSet" below using OperationType not LockType * With Static Partitions we assume that the query modifies exactly the partitions it locked. (not entirely * realistic since Update/Delete may have some predicate that filters out all records out of * some partition(s), but plausible). For DP, we acquire locks very wide (all known partitions), * but for most queries only a fraction will actually be updated. #addDynamicPartitions() tells * us exactly which ones were written to. Thus using this trick to kill a query early for * DP queries may be too restrictive. */ boolean isPartOfDynamicPartitionInsert = true; try { /** * checkLock() must be mutexed against any other checkLock to make sure 2 conflicting locks * are not granted by parallel checkLock() calls. */ handle = getMutexAPI().acquireLock(MUTEX_KEY.CheckLock.name()); List<LockInfo> locksBeingChecked = getLockInfoFromLockId(dbConn, extLockId);//being acquired now response.setLockid(extLockId); LOG.debug("checkLock(): Setting savepoint. extLockId=" + JavaUtils.lockIdToString(extLockId)); Savepoint save = dbConn.setSavepoint(); StringBuilder query = new StringBuilder( "select hl_lock_ext_id, " + "hl_lock_int_id, hl_db, hl_table, hl_partition, hl_lock_state, " + "hl_lock_type, hl_txnid from HIVE_LOCKS where hl_db in ("); Set<String> strings = new HashSet<String>(locksBeingChecked.size()); //This the set of entities that the statement represented by extLockId wants to update List<LockInfo> writeSet = new ArrayList<>(); for (LockInfo info : locksBeingChecked) { strings.add(info.db); if (!isPartOfDynamicPartitionInsert && info.type == LockType.SHARED_WRITE) { writeSet.add(info); } } if (!writeSet.isEmpty()) { if (writeSet.get(0).txnId == 0) { //Write operation always start a txn throw new IllegalStateException( "Found Write lock for " + JavaUtils.lockIdToString(extLockId) + " but no txnid"); } stmt = dbConn.createStatement(); StringBuilder sb = new StringBuilder( " ws_database, ws_table, ws_partition, " + "ws_txnid, ws_commit_id " + "from WRITE_SET where ws_commit_id >= " + writeSet.get(0).txnId + " and (");//see commitTxn() for more info on this inequality for (LockInfo info : writeSet) { sb.append("(ws_database = ").append(quoteString(info.db)).append(" and ws_table = ") .append(quoteString(info.table)).append(" and ws_partition ") .append(info.partition == null ? "is null" : "= " + quoteString(info.partition)) .append(") or "); } sb.setLength(sb.length() - 4);//nuke trailing " or " sb.append(")"); //1 row is sufficient to know we have to kill the query rs = stmt.executeQuery(sqlGenerator.addLimitClause(1, sb.toString())); if (rs.next()) { /** * if here, it means we found an already committed txn which overlaps with the current one and * it updated the same resource the current txn wants to update. By First-committer-wins * rule, current txn will not be allowed to commit so may as well kill it now; This is just an * optimization to prevent wasting cluster resources to run a query which is known to be DOA. * {@link #commitTxn(CommitTxnRequest)} has the primary responsibility to ensure this. * checkLock() runs at READ_COMMITTED so you could have another (Hive) txn running commitTxn() * in parallel and thus writing to WRITE_SET. commitTxn() logic is properly mutexed to ensure * that we don't "miss" any WW conflicts. We could've mutexed the checkLock() and commitTxn() * as well but this reduces concurrency for very little gain. * Note that update/delete (which runs as dynamic partition insert) acquires a lock on the table, * but WRITE_SET has entries for actual partitions updated. Thus this optimization will "miss" * the WW conflict but it will be caught in commitTxn() where actual partitions written are known. * This is OK since we want 2 concurrent updates that update different sets of partitions to both commit. */ String resourceName = rs.getString(1) + '/' + rs.getString(2); String partName = rs.getString(3); if (partName != null) { resourceName += '/' + partName; } String msg = "Aborting " + JavaUtils.txnIdToString(writeSet.get(0).txnId) + " since a concurrent committed transaction [" + JavaUtils.txnIdToString(rs.getLong(4)) + "," + rs.getLong(5) + "] has already updated resouce '" + resourceName + "'"; LOG.info(msg); if (abortTxns(dbConn, Collections.singletonList(writeSet.get(0).txnId), true) != 1) { throw new IllegalStateException(msg + " FAILED!"); } dbConn.commit(); throw new TxnAbortedException(msg); } close(rs, stmt, null); } boolean first = true; for (String s : strings) { if (first) first = false; else query.append(", "); query.append('\''); query.append(s); query.append('\''); } query.append(")"); // If any of the table requests are null, then I need to pull all the // table locks for this db. boolean sawNull = false; strings.clear(); for (LockInfo info : locksBeingChecked) { if (info.table == null) { sawNull = true; break; } else { strings.add(info.table); } } if (!sawNull) { query.append(" and (hl_table is null or hl_table in("); first = true; for (String s : strings) { if (first) first = false; else query.append(", "); query.append('\''); query.append(s); query.append('\''); } query.append("))"); // If any of the partition requests are null, then I need to pull all // partition locks for this table. sawNull = false; strings.clear(); for (LockInfo info : locksBeingChecked) { if (info.partition == null) { sawNull = true; break; } else { strings.add(info.partition); } } if (!sawNull) { query.append(" and (hl_partition is null or hl_partition in("); first = true; for (String s : strings) { if (first) first = false; else query.append(", "); query.append('\''); query.append(s); query.append('\''); } query.append("))"); } } query.append(" and hl_lock_ext_id <= ").append(extLockId); LOG.debug("Going to execute query <" + query.toString() + ">"); stmt = dbConn.createStatement(); rs = stmt.executeQuery(query.toString()); SortedSet<LockInfo> lockSet = new TreeSet<LockInfo>(new LockInfoComparator()); while (rs.next()) { lockSet.add(new LockInfo(rs)); } // Turn the tree set into an array so we can move back and forth easily // in it. LockInfo[] locks = lockSet.toArray(new LockInfo[lockSet.size()]); if (LOG.isTraceEnabled()) { LOG.trace("Locks to check(full): "); for (LockInfo info : locks) { LOG.trace(" " + info); } } for (LockInfo info : locksBeingChecked) { // Find the lock record we're checking int index = -1; for (int i = 0; i < locks.length; i++) { if (locks[i].equals(info)) { index = i; break; } } // If we didn't find the lock, then it must not be in the table if (index == -1) { LOG.debug("Going to rollback"); dbConn.rollback(); throw new MetaException( "How did we get here, we heartbeated our lock before we started! ( " + info + ")"); } // If we've found it and it's already been marked acquired, // then just look at the other locks. if (locks[index].state == LockState.ACQUIRED) { /**this is what makes this method @SafeToRetry*/ continue; } // Look at everything in front of this lock to see if it should block // it or not. boolean acquired = false; for (int i = index - 1; i >= 0; i--) { // Check if we're operating on the same database, if not, move on if (!locks[index].db.equals(locks[i].db)) { continue; } // If table is null on either of these, then they are claiming to // lock the whole database and we need to check it. Otherwise, // check if they are operating on the same table, if not, move on. if (locks[index].table != null && locks[i].table != null && !locks[index].table.equals(locks[i].table)) { continue; } // If partition is null on either of these, then they are claiming to // lock the whole table and we need to check it. Otherwise, // check if they are operating on the same partition, if not, move on. if (locks[index].partition != null && locks[i].partition != null && !locks[index].partition.equals(locks[i].partition)) { continue; } // We've found something that matches what we're trying to lock, // so figure out if we can lock it too. LockAction lockAction = jumpTable.get(locks[index].type).get(locks[i].type).get(locks[i].state); LOG.debug("desired Lock: " + info + " checked Lock: " + locks[i] + " action: " + lockAction); switch (lockAction) { case WAIT: if (!ignoreConflict(info, locks[i])) { /*we acquire all locks for a given query atomically; if 1 blocks, all go into (remain) in * Waiting state. wait() will undo any 'acquire()' which may have happened as part of * this (metastore db) transaction and then we record which lock blocked the lock * we were testing ('info').*/ wait(dbConn, save); String sqlText = "update HIVE_LOCKS" + " set HL_BLOCKEDBY_EXT_ID=" + locks[i].extLockId + ", HL_BLOCKEDBY_INT_ID=" + locks[i].intLockId + " where HL_LOCK_EXT_ID=" + info.extLockId + " and HL_LOCK_INT_ID=" + info.intLockId; LOG.debug("Executing sql: " + sqlText); int updCnt = stmt.executeUpdate(sqlText); if (updCnt != 1) { shouldNeverHappen(info.txnId, info.extLockId, info.intLockId); } LOG.debug("Going to commit"); dbConn.commit(); response.setState(LockState.WAITING); LOG.debug("Lock(" + info + ") waiting for Lock(" + locks[i] + ")"); return response; } //fall through to ACQUIRE case ACQUIRE: acquire(dbConn, stmt, extLockId, info); acquired = true; break; case KEEP_LOOKING: continue; } if (acquired) break; // We've acquired this lock component, // so get out of the loop and look at the next component. } // If we've arrived here and we have not already acquired, it means there's nothing in the // way of the lock, so acquire the lock. if (!acquired) acquire(dbConn, stmt, extLockId, info); } // We acquired all of the locks, so commit and return acquired. LOG.debug("Going to commit"); dbConn.commit(); response.setState(LockState.ACQUIRED); } finally { close(rs, stmt, null); if (handle != null) { handle.releaseLocks(); } } return response; } /** * the {@link #jumpTable} only deals with LockState/LockType. In some cases it's not * sufficient. For example, an EXCLUSIVE lock on partition should prevent SHARED_READ * on the table, but there is no reason for EXCLUSIVE on a table to prevent SHARED_READ * on a database. Similarly, EXCLUSIVE on a partition should not conflict with SHARED_READ on * a database. (SHARED_READ is usually acquired on a database to make sure it's not dropped * while some operation is performed on that db (e.g. show tables, created table, etc) * EXCLUSIVE on an object may mean it's being dropped or overwritten (for non-acid tables, * an Insert uses EXCLUSIVE as well)). */ private boolean ignoreConflict(LockInfo desiredLock, LockInfo existingLock) { return ((desiredLock.isDbLock() && desiredLock.type == LockType.SHARED_READ && existingLock.isTableLock() && existingLock.type == LockType.EXCLUSIVE) || (existingLock.isDbLock() && existingLock.type == LockType.SHARED_READ && desiredLock.isTableLock() && desiredLock.type == LockType.EXCLUSIVE) || (desiredLock.isDbLock() && desiredLock.type == LockType.SHARED_READ && existingLock.isPartitionLock() && existingLock.type == LockType.EXCLUSIVE) || (existingLock.isDbLock() && existingLock.type == LockType.SHARED_READ && desiredLock.isPartitionLock() && desiredLock.type == LockType.EXCLUSIVE)) || //different locks from same txn should not conflict with each other (desiredLock.txnId != 0 && desiredLock.txnId == existingLock.txnId) || //txnId=0 means it's a select or IUD which does not write to ACID table, e.g //insert overwrite table T partition(p=1) select a,b from T and autoCommit=true // todo: fix comment as of HIVE-14988 (desiredLock.txnId == 0 && desiredLock.extLockId == existingLock.extLockId); } private void wait(Connection dbConn, Savepoint save) throws SQLException { // Need to rollback because we did a select that acquired locks but we didn't // actually update anything. Also, we may have locked some locks as // acquired that we now want to not acquire. It's ok to rollback because // once we see one wait, we're done, we won't look for more. // Only rollback to savepoint because we want to commit our heartbeat // changes. LOG.debug("Going to rollback to savepoint"); dbConn.rollback(save); } private void acquire(Connection dbConn, Statement stmt, long extLockId, LockInfo lockInfo) throws SQLException, NoSuchLockException, MetaException { long now = getDbTime(dbConn); String s = "update HIVE_LOCKS set hl_lock_state = '" + LOCK_ACQUIRED + "', " + //if lock is part of txn, heartbeat info is in txn record "hl_last_heartbeat = " + (isValidTxn(lockInfo.txnId) ? 0 : now) + ", hl_acquired_at = " + now + ",HL_BLOCKEDBY_EXT_ID=NULL,HL_BLOCKEDBY_INT_ID=null" + " where hl_lock_ext_id = " + extLockId + " and hl_lock_int_id = " + lockInfo.intLockId; LOG.debug("Going to execute update <" + s + ">"); int rc = stmt.executeUpdate(s); if (rc < 1) { LOG.debug("Going to rollback"); dbConn.rollback(); throw new NoSuchLockException("No such lock: (" + JavaUtils.lockIdToString(extLockId) + "," + +lockInfo.intLockId + ") " + JavaUtils.txnIdToString(lockInfo.txnId)); } // We update the database, but we don't commit because there may be other // locks together with this, and we only want to acquire one if we can // acquire all. } /** * Heartbeats on the lock table. This commits, so do not enter it with any state. * Should not be called on a lock that belongs to transaction. */ private void heartbeatLock(Connection dbConn, long extLockId) throws NoSuchLockException, SQLException, MetaException { // If the lock id is 0, then there are no locks in this heartbeat if (extLockId == 0) return; Statement stmt = null; try { stmt = dbConn.createStatement(); long now = getDbTime(dbConn); String s = "update HIVE_LOCKS set hl_last_heartbeat = " + now + " where hl_lock_ext_id = " + extLockId; LOG.debug("Going to execute update <" + s + ">"); int rc = stmt.executeUpdate(s); if (rc < 1) { LOG.debug("Going to rollback"); dbConn.rollback(); throw new NoSuchLockException("No such lock: " + JavaUtils.lockIdToString(extLockId)); } LOG.debug("Going to commit"); dbConn.commit(); } finally { closeStmt(stmt); } } // Heartbeats on the txn table. This commits, so do not enter it with any state private void heartbeatTxn(Connection dbConn, long txnid) throws NoSuchTxnException, TxnAbortedException, SQLException, MetaException { // If the txnid is 0, then there are no transactions in this heartbeat if (txnid == 0) return; Statement stmt = null; try { stmt = dbConn.createStatement(); long now = getDbTime(dbConn); String s = "update TXNS set txn_last_heartbeat = " + now + " where txn_id = " + txnid + " and txn_state = '" + TXN_OPEN + "'"; LOG.debug("Going to execute update <" + s + ">"); int rc = stmt.executeUpdate(s); if (rc < 1) { ensureValidTxn(dbConn, txnid, stmt); // This should now throw some useful exception. LOG.warn("Can neither heartbeat txn nor confirm it as invalid."); dbConn.rollback(); throw new NoSuchTxnException("No such txn: " + txnid); } LOG.debug("Going to commit"); dbConn.commit(); } finally { closeStmt(stmt); } } /** * Returns the state of the transaction iff it's able to determine it. Some cases where it cannot: * 1. txnid was Aborted/Committed and then GC'd (compacted) * 2. txnid was committed but it didn't modify anything (nothing in COMPLETED_TXN_COMPONENTS) */ private TxnStatus findTxnState(long txnid, Statement stmt) throws SQLException, MetaException { String s = "select txn_state from TXNS where txn_id = " + txnid; LOG.debug("Going to execute query <" + s + ">"); try (ResultSet rs = stmt.executeQuery(s)) { if (!rs.next()) { s = sqlGenerator.addLimitClause(1, "1 from COMPLETED_TXN_COMPONENTS where CTC_TXNID = " + txnid); LOG.debug("Going to execute query <" + s + ">"); try (ResultSet rs2 = stmt.executeQuery(s)) { if (rs2.next()) { return TxnStatus.COMMITTED; } } // could also check WRITE_SET but that seems overkill return TxnStatus.UNKNOWN; } char txnState = rs.getString(1).charAt(0); if (txnState == TXN_ABORTED) { return TxnStatus.ABORTED; } assert txnState == TXN_OPEN : "we found it in TXNS but it's not ABORTED, so must be OPEN"; } return TxnStatus.OPEN; } /** * Used to raise an informative error when the caller expected a txn in a particular TxnStatus * but found it in some other status */ private static void raiseTxnUnexpectedState(TxnStatus actualStatus, long txnid) throws NoSuchTxnException, TxnAbortedException { switch (actualStatus) { case ABORTED: throw new TxnAbortedException("Transaction " + JavaUtils.txnIdToString(txnid) + " already aborted"); case COMMITTED: throw new NoSuchTxnException( "Transaction " + JavaUtils.txnIdToString(txnid) + " is already committed."); case UNKNOWN: throw new NoSuchTxnException("No such transaction " + JavaUtils.txnIdToString(txnid)); case OPEN: throw new NoSuchTxnException(JavaUtils.txnIdToString(txnid) + " is " + TxnStatus.OPEN); default: throw new IllegalArgumentException("Unknown TxnStatus " + actualStatus); } } /** * Returns the state of the transaction with {@code txnid} or throws if {@code raiseError} is true. */ private static void ensureValidTxn(Connection dbConn, long txnid, Statement stmt) throws SQLException, NoSuchTxnException, TxnAbortedException { // We need to check whether this transaction is valid and open String s = "select txn_state from TXNS where txn_id = " + txnid; LOG.debug("Going to execute query <" + s + ">"); try (ResultSet rs = stmt.executeQuery(s)) { if (!rs.next()) { // todo: add LIMIT 1 instead of count - should be more efficient s = "select count(*) from COMPLETED_TXN_COMPONENTS where CTC_TXNID = " + txnid; try (ResultSet rs2 = stmt.executeQuery(s)) { // todo: strictly speaking you can commit an empty txn, thus 2nd conjunct is wrong but // only // possible for for multi-stmt txns boolean alreadyCommitted = rs2.next() && rs2.getInt(1) > 0; LOG.debug("Going to rollback"); rollbackDBConn(dbConn); if (alreadyCommitted) { // makes the message more informative - helps to find bugs in client code throw new NoSuchTxnException( "Transaction " + JavaUtils.txnIdToString(txnid) + " is already committed."); } throw new NoSuchTxnException("No such transaction " + JavaUtils.txnIdToString(txnid)); } } if (rs.getString(1).charAt(0) == TXN_ABORTED) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); throw new TxnAbortedException("Transaction " + JavaUtils.txnIdToString(txnid) + " already aborted");// todo: add time of abort, which is not currently tracked. // Requires schema change } } } private LockInfo getTxnIdFromLockId(Connection dbConn, long extLockId) throws NoSuchLockException, MetaException, SQLException { Statement stmt = null; ResultSet rs = null; try { stmt = dbConn.createStatement(); String s = "select hl_lock_ext_id, hl_lock_int_id, hl_db, hl_table, " + "hl_partition, hl_lock_state, hl_lock_type, hl_txnid from HIVE_LOCKS where " + "hl_lock_ext_id = " + extLockId; LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { return null; } LockInfo info = new LockInfo(rs); LOG.debug("getTxnIdFromLockId(" + extLockId + ") Return " + JavaUtils.txnIdToString(info.txnId)); return info; } finally { close(rs); closeStmt(stmt); } } // NEVER call this function without first calling heartbeat(long, long) private List<LockInfo> getLockInfoFromLockId(Connection dbConn, long extLockId) throws NoSuchLockException, MetaException, SQLException { Statement stmt = null; try { stmt = dbConn.createStatement(); String s = "select hl_lock_ext_id, hl_lock_int_id, hl_db, hl_table, " + "hl_partition, hl_lock_state, hl_lock_type, hl_txnid from HIVE_LOCKS where " + "hl_lock_ext_id = " + extLockId; LOG.debug("Going to execute query <" + s + ">"); ResultSet rs = stmt.executeQuery(s); boolean sawAtLeastOne = false; List<LockInfo> ourLockInfo = new ArrayList<LockInfo>(); while (rs.next()) { ourLockInfo.add(new LockInfo(rs)); sawAtLeastOne = true; } if (!sawAtLeastOne) { throw new MetaException("This should never happen! We already " + "checked the lock(" + JavaUtils.lockIdToString(extLockId) + ") existed but now we can't find it!"); } return ourLockInfo; } finally { closeStmt(stmt); } } // Clean time out locks from the database not associated with a transactions, i.e. locks // for read-only autoCommit=true statements. This does a commit, // and thus should be done before any calls to heartbeat that will leave // open transactions. private void timeOutLocks(Connection dbConn, long now) { Statement stmt = null; ResultSet rs = null; try { stmt = dbConn.createStatement(); long maxHeartbeatTime = now - timeout; //doing a SELECT first is less efficient but makes it easier to debug things String s = "select distinct hl_lock_ext_id from HIVE_LOCKS where hl_last_heartbeat < " + maxHeartbeatTime + " and hl_txnid = 0";//when txnid is <> 0, the lock is //associated with a txn and is handled by performTimeOuts() //want to avoid expiring locks for a txn w/o expiring the txn itself List<Long> extLockIDs = new ArrayList<>(); rs = stmt.executeQuery(s); while (rs.next()) { extLockIDs.add(rs.getLong(1)); } rs.close(); dbConn.commit(); if (extLockIDs.size() <= 0) { return; } List<String> queries = new ArrayList<String>(); StringBuilder prefix = new StringBuilder(); StringBuilder suffix = new StringBuilder(); //include same hl_last_heartbeat condition in case someone heartbeated since the select prefix.append("delete from HIVE_LOCKS where hl_last_heartbeat < "); prefix.append(maxHeartbeatTime); prefix.append(" and hl_txnid = 0 and "); suffix.append(""); TxnUtils.buildQueryWithINClause(conf, queries, prefix, suffix, extLockIDs, "hl_lock_ext_id", true, false); int deletedLocks = 0; for (String query : queries) { LOG.debug("Removing expired locks via: " + query); deletedLocks += stmt.executeUpdate(query); } if (deletedLocks > 0) { Collections.sort(extLockIDs);//easier to read logs LOG.info("Deleted " + deletedLocks + " int locks from HIVE_LOCKS due to timeout (" + "HL_LOCK_EXT_ID list: " + extLockIDs + ") maxHeartbeatTime=" + maxHeartbeatTime); } LOG.debug("Going to commit"); dbConn.commit(); } catch (SQLException ex) { LOG.error("Failed to purge timedout locks due to: " + getMessage(ex), ex); } catch (Exception ex) { LOG.error("Failed to purge timedout locks due to: " + ex.getMessage(), ex); } finally { close(rs); closeStmt(stmt); } } /** * Isolation Level Notes * Plain: RC is OK * This will find transactions that have timed out and abort them. * Will also delete locks which are not associated with a transaction and have timed out * Tries to keep transactions (against metastore db) small to reduce lock contention. */ @RetrySemantics.Idempotent public void performTimeOuts() { Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); //We currently commit after selecting the TXNS to abort. So whether SERIALIZABLE //READ_COMMITTED, the effect is the same. We could use FOR UPDATE on Select from TXNS //and do the whole performTimeOuts() in a single huge transaction, but the only benefit //would be to make sure someone cannot heartbeat one of these txns at the same time. //The attempt to heartbeat would block and fail immediately after it's unblocked. //With current (RC + multiple txns) implementation it is possible for someone to send //heartbeat at the very end of the expire interval, and just after the Select from TXNS //is made, in which case heartbeat will succeed but txn will still be Aborted. //Solving this corner case is not worth the perf penalty. The client should heartbeat in a //timely way. long now = getDbTime(dbConn); timeOutLocks(dbConn, now); while (true) { stmt = dbConn.createStatement(); String s = " txn_id from TXNS where txn_state = '" + TXN_OPEN + "' and txn_last_heartbeat < " + (now - timeout); //safety valve for extreme cases s = sqlGenerator.addLimitClause(10 * TIMED_OUT_TXN_ABORT_BATCH_SIZE, s); LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { return;//no more timedout txns } List<List<Long>> timedOutTxns = new ArrayList<>(); List<Long> currentBatch = new ArrayList<>(TIMED_OUT_TXN_ABORT_BATCH_SIZE); timedOutTxns.add(currentBatch); do { if (currentBatch.size() == TIMED_OUT_TXN_ABORT_BATCH_SIZE) { currentBatch = new ArrayList<>(TIMED_OUT_TXN_ABORT_BATCH_SIZE); timedOutTxns.add(currentBatch); } currentBatch.add(rs.getLong(1)); } while (rs.next()); dbConn.commit(); close(rs, stmt, null); int numTxnsAborted = 0; for (List<Long> batchToAbort : timedOutTxns) { if (abortTxns(dbConn, batchToAbort, now - timeout, true) == batchToAbort.size()) { dbConn.commit(); numTxnsAborted += batchToAbort.size(); //todo: add TXNS.COMMENT filed and set it to 'aborted by system due to timeout' Collections.sort(batchToAbort);//easier to read logs LOG.info("Aborted the following transactions due to timeout: " + batchToAbort.toString()); } else { //could not abort all txns in this batch - this may happen because in parallel with this //operation there was activity on one of the txns in this batch (commit/abort/heartbeat) //This is not likely but may happen if client experiences long pause between heartbeats or //unusually long/extreme pauses between heartbeat() calls and other logic in checkLock(), //lock(), etc. dbConn.rollback(); } } LOG.info("Aborted " + numTxnsAborted + " transactions due to timeout"); } } catch (SQLException ex) { LOG.warn("Aborting timedout transactions failed due to " + getMessage(ex), ex); } catch (MetaException e) { LOG.warn("Aborting timedout transactions failed due to " + e.getMessage(), e); } finally { close(rs, stmt, dbConn); } } @Override @RetrySemantics.ReadOnly public void countOpenTxns() throws MetaException { Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { try { dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED); stmt = dbConn.createStatement(); String s = "select count(*) from TXNS where txn_state = '" + TXN_OPEN + "'"; LOG.debug("Going to execute query <" + s + ">"); rs = stmt.executeQuery(s); if (!rs.next()) { LOG.error("Transaction database not properly configured, " + "can't find txn_state from TXNS."); } else { numOpenTxns = rs.getLong(1); } } catch (SQLException e) { LOG.debug("Going to rollback"); rollbackDBConn(dbConn); LOG.info("Failed to update number of open transactions"); checkRetryable(dbConn, e, "countOpenTxns()"); } finally { close(rs, stmt, dbConn); } } catch (RetryException e) { countOpenTxns(); } } private static synchronized DataSource setupJdbcConnectionPool(HiveConf conf, int maxPoolSize, long getConnectionTimeoutMs) throws SQLException { String driverUrl = HiveConf.getVar(conf, HiveConf.ConfVars.METASTORECONNECTURLKEY); String user = getMetastoreJdbcUser(conf); String passwd = getMetastoreJdbcPasswd(conf); String connectionPooler = conf.getVar(HiveConf.ConfVars.METASTORE_CONNECTION_POOLING_TYPE).toLowerCase(); if ("bonecp".equals(connectionPooler)) { BoneCPConfig config = new BoneCPConfig(); config.setJdbcUrl(driverUrl); //if we are waiting for connection for a long time, something is really wrong //better raise an error than hang forever //see DefaultConnectionStrategy.getConnectionInternal() config.setConnectionTimeoutInMs(getConnectionTimeoutMs); config.setMaxConnectionsPerPartition(maxPoolSize); config.setPartitionCount(1); config.setUser(user); config.setPassword(passwd); doRetryOnConnPool = true; // Enable retries to work around BONECP bug. return new BoneCPDataSource(config); } else if ("dbcp".equals(connectionPooler)) { GenericObjectPool objectPool = new GenericObjectPool(); //https://commons.apache.org/proper/commons-pool/api-1.6/org/apache/commons/pool/impl/GenericObjectPool.html#setMaxActive(int) objectPool.setMaxActive(maxPoolSize); objectPool.setMaxWait(getConnectionTimeoutMs); ConnectionFactory connFactory = new DriverManagerConnectionFactory(driverUrl, user, passwd); // This doesn't get used, but it's still necessary, see // http://svn.apache.org/viewvc/commons/proper/dbcp/branches/DBCP_1_4_x_BRANCH/doc/ManualPoolingDataSourceExample.java?view=markup PoolableConnectionFactory poolConnFactory = new PoolableConnectionFactory(connFactory, objectPool, null, null, false, true); return new PoolingDataSource(objectPool); } else if ("hikaricp".equals(connectionPooler)) { HikariConfig config = new HikariConfig(); config.setMaximumPoolSize(maxPoolSize); config.setJdbcUrl(driverUrl); config.setUsername(user); config.setPassword(passwd); //https://github.com/brettwooldridge/HikariCP config.setConnectionTimeout(getConnectionTimeoutMs); return new HikariDataSource(config); } else if ("none".equals(connectionPooler)) { LOG.info("Choosing not to pool JDBC connections"); return new NoPoolConnectionPool(conf); } else { throw new RuntimeException("Unknown JDBC connection pooling " + connectionPooler); } } private static synchronized void buildJumpTable() { if (jumpTable != null) return; jumpTable = new HashMap<LockType, Map<LockType, Map<LockState, LockAction>>>(3); // SR: Lock we are trying to acquire is shared read Map<LockType, Map<LockState, LockAction>> m = new HashMap<LockType, Map<LockState, LockAction>>(3); jumpTable.put(LockType.SHARED_READ, m); // SR.SR: Lock we are examining is shared read Map<LockState, LockAction> m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_READ, m2); // SR.SR.acquired Lock we are examining is acquired; We can acquire // because two shared reads can acquire together and there must be // nothing in front of this one to prevent acquisition. m2.put(LockState.ACQUIRED, LockAction.ACQUIRE); // SR.SR.wait Lock we are examining is waiting. In this case we keep // looking, as it's possible that something in front is blocking it or // that the other locker hasn't checked yet and he could lock as well. m2.put(LockState.WAITING, LockAction.KEEP_LOOKING); // SR.SW: Lock we are examining is shared write m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_WRITE, m2); // SR.SW.acquired Lock we are examining is acquired; We can acquire // because a read can share with a write, and there must be // nothing in front of this one to prevent acquisition. m2.put(LockState.ACQUIRED, LockAction.ACQUIRE); // SR.SW.wait Lock we are examining is waiting. In this case we keep // looking, as it's possible that something in front is blocking it or // that the other locker hasn't checked yet and he could lock as well or // that something is blocking it that would not block a read. m2.put(LockState.WAITING, LockAction.KEEP_LOOKING); // SR.E: Lock we are examining is exclusive m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.EXCLUSIVE, m2); // No matter whether it has acquired or not, we cannot pass an exclusive. m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); // SW: Lock we are trying to acquire is shared write m = new HashMap<LockType, Map<LockState, LockAction>>(3); jumpTable.put(LockType.SHARED_WRITE, m); // SW.SR: Lock we are examining is shared read m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_READ, m2); // SW.SR.acquired Lock we are examining is acquired; We need to keep // looking, because there may or may not be another shared write in front // that would block us. m2.put(LockState.ACQUIRED, LockAction.KEEP_LOOKING); // SW.SR.wait Lock we are examining is waiting. In this case we keep // looking, as it's possible that something in front is blocking it or // that the other locker hasn't checked yet and he could lock as well. m2.put(LockState.WAITING, LockAction.KEEP_LOOKING); // SW.SW: Lock we are examining is shared write m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_WRITE, m2); // Regardless of acquired or waiting, one shared write cannot pass another. m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); // SW.E: Lock we are examining is exclusive m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.EXCLUSIVE, m2); // No matter whether it has acquired or not, we cannot pass an exclusive. m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); // E: Lock we are trying to acquire is exclusive m = new HashMap<LockType, Map<LockState, LockAction>>(3); jumpTable.put(LockType.EXCLUSIVE, m); // E.SR: Lock we are examining is shared read m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_READ, m2); // Exclusives can never pass m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); // E.SW: Lock we are examining is shared write m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.SHARED_WRITE, m2); // Exclusives can never pass m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); // E.E: Lock we are examining is exclusive m2 = new HashMap<LockState, LockAction>(2); m.put(LockType.EXCLUSIVE, m2); // No matter whether it has acquired or not, we cannot pass an exclusive. m2.put(LockState.ACQUIRED, LockAction.WAIT); m2.put(LockState.WAITING, LockAction.WAIT); } /** * Returns true if {@code ex} should be retried */ static boolean isRetryable(HiveConf conf, Exception ex) { if (ex instanceof SQLException) { SQLException sqlException = (SQLException) ex; if ("08S01".equalsIgnoreCase(sqlException.getSQLState())) { //in MSSQL this means Communication Link Failure return true; } if ("ORA-08176".equalsIgnoreCase(sqlException.getSQLState()) || sqlException.getMessage().contains("consistent read failure; rollback data not available")) { return true; } String regex = HiveConf.getVar(conf, HiveConf.ConfVars.HIVE_TXN_RETRYABLE_SQLEX_REGEX); if (regex != null && !regex.isEmpty()) { String[] patterns = regex.split(",(?=\\S)"); String message = getMessage((SQLException) ex); for (String p : patterns) { if (Pattern.matches(p, message)) { return true; } } } //see also https://issues.apache.org/jira/browse/HIVE-9938 } return false; } private boolean isDuplicateKeyError(SQLException ex) { switch (dbProduct) { case DERBY: if ("23505".equals(ex.getSQLState())) { return true; } break; case MYSQL: //https://dev.mysql.com/doc/refman/5.5/en/error-messages-server.html if ((ex.getErrorCode() == 1022 || ex.getErrorCode() == 1062 || ex.getErrorCode() == 1586) && "23000".equals(ex.getSQLState())) { return true; } break; case SQLSERVER: //2627 is unique constaint violation incl PK, 2601 - unique key if (ex.getErrorCode() == 2627 && "23000".equals(ex.getSQLState())) { return true; } break; case ORACLE: if (ex.getErrorCode() == 1 && "23000".equals(ex.getSQLState())) { return true; } break; case POSTGRES: //http://www.postgresql.org/docs/8.1/static/errcodes-appendix.html if ("23505".equals(ex.getSQLState())) { return true; } break; default: throw new IllegalArgumentException("Unexpected DB type: " + dbProduct + "; " + getMessage(ex)); } return false; } private static String getMessage(SQLException ex) { return ex.getMessage() + " (SQLState=" + ex.getSQLState() + ", ErrorCode=" + ex.getErrorCode() + ")"; } /** * Useful for building SQL strings * @param value may be {@code null} */ private static String valueOrNullLiteral(String value) { return value == null ? "null" : quoteString(value); } static String quoteString(String input) { return "'" + input + "'"; } static String quoteChar(char c) { return "'" + c + "'"; } static CompactionType dbCompactionType2ThriftType(char dbValue) { switch (dbValue) { case MAJOR_TYPE: return CompactionType.MAJOR; case MINOR_TYPE: return CompactionType.MINOR; default: LOG.warn("Unexpected compaction type " + dbValue); return null; } } static Character thriftCompactionType2DbType(CompactionType ct) { switch (ct) { case MAJOR: return MAJOR_TYPE; case MINOR: return MINOR_TYPE; default: LOG.warn("Unexpected compaction type " + ct); return null; } } /** * {@link #lockInternal()} and {@link #unlockInternal()} are used to serialize those operations that require * Select ... For Update to sequence operations properly. In practice that means when running * with Derby database. See more notes at class level. */ private void lockInternal() { if (dbProduct == DatabaseProduct.DERBY) { derbyLock.lock(); } } private void unlockInternal() { if (dbProduct == DatabaseProduct.DERBY) { derbyLock.unlock(); } } @Override @RetrySemantics.Idempotent public MutexAPI getMutexAPI() { return this; } @Override public LockHandle acquireLock(String key) throws MetaException { /** * The implementation here is a bit kludgey but done so that code exercised by unit tests * (which run against Derby which has no support for select for update) is as similar to * production code as possible. * In particular, with Derby we always run in a single process with a single metastore and * the absence of For Update is handled via a Semaphore. The later would strictly speaking * make the SQL statements below unnecessary (for Derby), but then they would not be tested. */ Connection dbConn = null; Statement stmt = null; ResultSet rs = null; try { try { String sqlStmt = sqlGenerator.addForUpdateClause( "select MT_COMMENT from AUX_TABLE where MT_KEY1=" + quoteString(key) + " and MT_KEY2=0"); lockInternal(); dbConn = getDbConn(Connection.TRANSACTION_READ_COMMITTED, connPoolMutex); stmt = dbConn.createStatement(); if (LOG.isDebugEnabled()) { LOG.debug("About to execute SQL: " + sqlStmt); } rs = stmt.executeQuery(sqlStmt); if (!rs.next()) { close(rs); try { stmt.executeUpdate( "insert into AUX_TABLE(MT_KEY1,MT_KEY2) values(" + quoteString(key) + ", 0)"); dbConn.commit(); } catch (SQLException ex) { if (!isDuplicateKeyError(ex)) { throw new RuntimeException( "Unable to lock " + quoteString(key) + " due to: " + getMessage(ex), ex); } //if here, it means a concrurrent acquireLock() inserted the 'key' //rollback is done for the benefit of Postgres which throws (SQLState=25P02, ErrorCode=0) if //you attempt any stmt in a txn which had an error. dbConn.rollback(); } rs = stmt.executeQuery(sqlStmt); if (!rs.next()) { throw new IllegalStateException( "Unable to lock " + quoteString(key) + ". Expected row in AUX_TABLE is missing."); } } Semaphore derbySemaphore = null; if (dbProduct == DatabaseProduct.DERBY) { derbyKey2Lock.putIfAbsent(key, new Semaphore(1)); derbySemaphore = derbyKey2Lock.get(key); derbySemaphore.acquire(); } LOG.debug(quoteString(key) + " locked by " + quoteString(TxnHandler.hostname)); //OK, so now we have a lock return new LockHandleImpl(dbConn, stmt, rs, key, derbySemaphore); } catch (SQLException ex) { rollbackDBConn(dbConn); close(rs, stmt, dbConn); checkRetryable(dbConn, ex, "acquireLock(" + key + ")"); throw new MetaException("Unable to lock " + quoteString(key) + " due to: " + getMessage(ex) + "; " + StringUtils.stringifyException(ex)); } catch (InterruptedException ex) { rollbackDBConn(dbConn); close(rs, stmt, dbConn); throw new MetaException("Unable to lock " + quoteString(key) + " due to: " + ex.getMessage() + StringUtils.stringifyException(ex)); } finally { unlockInternal(); } } catch (RetryException ex) { return acquireLock(key); } } public void acquireLock(String key, LockHandle handle) { //the idea is that this will use LockHandle.dbConn throw new NotImplementedException(); } private static final class LockHandleImpl implements LockHandle { private final Connection dbConn; private final Statement stmt; private final ResultSet rs; private final Semaphore derbySemaphore; private final List<String> keys = new ArrayList<>(); LockHandleImpl(Connection conn, Statement stmt, ResultSet rs, String key, Semaphore derbySemaphore) { this.dbConn = conn; this.stmt = stmt; this.rs = rs; this.derbySemaphore = derbySemaphore; if (derbySemaphore != null) { //oterwise it may later release permit acquired by someone else assert derbySemaphore.availablePermits() == 0 : "Expected locked Semaphore"; } keys.add(key); } void addKey(String key) { //keys.add(key); //would need a list of (stmt,rs) pairs - 1 for each key throw new NotImplementedException(); } @Override public void releaseLocks() { rollbackDBConn(dbConn); close(rs, stmt, dbConn); if (derbySemaphore != null) { derbySemaphore.release(); } for (String key : keys) { LOG.debug(quoteString(key) + " unlocked by " + quoteString(TxnHandler.hostname)); } } } /** * Helper class that generates SQL queries with syntax specific to target DB * todo: why throw MetaException? */ @VisibleForTesting static final class SQLGenerator { private final DatabaseProduct dbProduct; private final HiveConf conf; SQLGenerator(DatabaseProduct dbProduct, HiveConf conf) { this.dbProduct = dbProduct; this.conf = conf; } /** * Genereates "Insert into T(a,b,c) values(1,2,'f'),(3,4,'c')" for appropriate DB * @param tblColumns e.g. "T(a,b,c)" * @param rows e.g. list of Strings like 3,4,'d' * @return fully formed INSERT INTO ... statements */ List<String> createInsertValuesStmt(String tblColumns, List<String> rows) { if (rows == null || rows.size() == 0) { return Collections.emptyList(); } List<String> insertStmts = new ArrayList<>(); StringBuilder sb = new StringBuilder(); switch (dbProduct) { case ORACLE: if (rows.size() > 1) { //http://www.oratable.com/oracle-insert-all/ //https://livesql.oracle.com/apex/livesql/file/content_BM1LJQ87M5CNIOKPOWPV6ZGR3.html for (int numRows = 0; numRows < rows.size(); numRows++) { if (numRows % conf.getIntVar( HiveConf.ConfVars.METASTORE_DIRECT_SQL_MAX_ELEMENTS_VALUES_CLAUSE) == 0) { if (numRows > 0) { sb.append(" select * from dual"); insertStmts.add(sb.toString()); } sb.setLength(0); sb.append("insert all "); } sb.append("into ").append(tblColumns).append(" values(").append(rows.get(numRows)) .append(") "); } sb.append("select * from dual"); insertStmts.add(sb.toString()); return insertStmts; } //fall through case DERBY: case MYSQL: case POSTGRES: case SQLSERVER: for (int numRows = 0; numRows < rows.size(); numRows++) { if (numRows % conf .getIntVar(HiveConf.ConfVars.METASTORE_DIRECT_SQL_MAX_ELEMENTS_VALUES_CLAUSE) == 0) { if (numRows > 0) { insertStmts.add(sb.substring(0, sb.length() - 1));//exclude trailing comma } sb.setLength(0); sb.append("insert into ").append(tblColumns).append(" values"); } sb.append('(').append(rows.get(numRows)).append("),"); } insertStmts.add(sb.substring(0, sb.length() - 1));//exclude trailing comma return insertStmts; default: String msg = "Unrecognized database product name <" + dbProduct + ">"; LOG.error(msg); throw new IllegalStateException(msg); } } /** * Given a {@code selectStatement}, decorated it with FOR UPDATE or semantically equivalent * construct. If the DB doesn't support, return original select. */ String addForUpdateClause(String selectStatement) throws MetaException { switch (dbProduct) { case DERBY: //https://db.apache.org/derby/docs/10.1/ref/rrefsqlj31783.html //sadly in Derby, FOR UPDATE doesn't meant what it should return selectStatement; case MYSQL: //http://dev.mysql.com/doc/refman/5.7/en/select.html case ORACLE: //https://docs.oracle.com/cd/E17952_01/refman-5.6-en/select.html case POSTGRES: //http://www.postgresql.org/docs/9.0/static/sql-select.html return selectStatement + " for update"; case SQLSERVER: //https://msdn.microsoft.com/en-us/library/ms189499.aspx //https://msdn.microsoft.com/en-us/library/ms187373.aspx String modifier = " with (updlock)"; int wherePos = selectStatement.toUpperCase().indexOf(" WHERE "); if (wherePos < 0) { return selectStatement + modifier; } return selectStatement.substring(0, wherePos) + modifier + selectStatement.substring(wherePos, selectStatement.length()); default: String msg = "Unrecognized database product name <" + dbProduct + ">"; LOG.error(msg); throw new MetaException(msg); } } /** * Suppose you have a query "select a,b from T" and you want to limit the result set * to the first 5 rows. The mechanism to do that differs in different DBs. * Make {@code noSelectsqlQuery} to be "a,b from T" and this method will return the * appropriately modified row limiting query. * * Note that if {@code noSelectsqlQuery} contains a join, you must make sure that * all columns are unique for Oracle. */ private String addLimitClause(int numRows, String noSelectsqlQuery) throws MetaException { switch (dbProduct) { case DERBY: //http://db.apache.org/derby/docs/10.7/ref/rrefsqljoffsetfetch.html return "select " + noSelectsqlQuery + " fetch first " + numRows + " rows only"; case MYSQL: //http://www.postgresql.org/docs/7.3/static/queries-limit.html case POSTGRES: //https://dev.mysql.com/doc/refman/5.0/en/select.html return "select " + noSelectsqlQuery + " limit " + numRows; case ORACLE: //newer versions (12c and later) support OFFSET/FETCH return "select * from (select " + noSelectsqlQuery + ") where rownum <= " + numRows; case SQLSERVER: //newer versions (2012 and later) support OFFSET/FETCH //https://msdn.microsoft.com/en-us/library/ms189463.aspx return "select TOP(" + numRows + ") " + noSelectsqlQuery; default: String msg = "Unrecognized database product name <" + dbProduct + ">"; LOG.error(msg); throw new MetaException(msg); } } } private static String getMetastoreJdbcUser(HiveConf conf) { return conf.getVar(HiveConf.ConfVars.METASTORE_CONNECTION_USER_NAME); } private static String getMetastoreJdbcPasswd(HiveConf conf) throws SQLException { try { return ShimLoader.getHadoopShims().getPassword(conf, HiveConf.ConfVars.METASTOREPWD.varname); } catch (IOException err) { throw new SQLException("Error getting metastore password", err); } } private static class NoPoolConnectionPool implements DataSource { // Note that this depends on the fact that no-one in this class calls anything but // getConnection. If you want to use any of the Logger or wrap calls you'll have to // implement them. private final HiveConf conf; private Driver driver; private String connString; private String user; private String passwd; public NoPoolConnectionPool(HiveConf conf) { this.conf = conf; } @Override public Connection getConnection() throws SQLException { if (user == null) { user = getMetastoreJdbcUser(conf); passwd = getMetastoreJdbcPasswd(conf); } return getConnection(user, passwd); } @Override public Connection getConnection(String username, String password) throws SQLException { // Find the JDBC driver if (driver == null) { String driverName = conf.getVar(HiveConf.ConfVars.METASTORE_CONNECTION_DRIVER); if (driverName == null || driverName.equals("")) { String msg = "JDBC driver for transaction db not set in configuration " + "file, need to set " + HiveConf.ConfVars.METASTORE_CONNECTION_DRIVER.varname; LOG.error(msg); throw new RuntimeException(msg); } try { LOG.info("Going to load JDBC driver " + driverName); driver = (Driver) Class.forName(driverName).newInstance(); } catch (InstantiationException e) { throw new RuntimeException("Unable to instantiate driver " + driverName + ", " + e.getMessage(), e); } catch (IllegalAccessException e) { throw new RuntimeException("Unable to access driver " + driverName + ", " + e.getMessage(), e); } catch (ClassNotFoundException e) { throw new RuntimeException("Unable to find driver " + driverName + ", " + e.getMessage(), e); } connString = conf.getVar(HiveConf.ConfVars.METASTORECONNECTURLKEY); } try { LOG.info("Connecting to transaction db with connection string " + connString); Properties connectionProps = new Properties(); connectionProps.setProperty("user", username); connectionProps.setProperty("password", password); Connection conn = driver.connect(connString, connectionProps); conn.setAutoCommit(false); return conn; } catch (SQLException e) { throw new RuntimeException( "Unable to connect to transaction manager using " + connString + ", " + e.getMessage(), e); } } @Override public PrintWriter getLogWriter() throws SQLException { throw new UnsupportedOperationException(); } @Override public void setLogWriter(PrintWriter out) throws SQLException { throw new UnsupportedOperationException(); } @Override public void setLoginTimeout(int seconds) throws SQLException { throw new UnsupportedOperationException(); } @Override public int getLoginTimeout() throws SQLException { throw new UnsupportedOperationException(); } @Override public java.util.logging.Logger getParentLogger() throws SQLFeatureNotSupportedException { throw new UnsupportedOperationException(); } @Override public <T> T unwrap(Class<T> iface) throws SQLException { throw new UnsupportedOperationException(); } @Override public boolean isWrapperFor(Class<?> iface) throws SQLException { throw new UnsupportedOperationException(); } }; }