use of org.apache.hadoop.hive.metastore.txn.TxnStore in project hive by apache.
the class TestReplicationScenariosAcidTables method testTargetDbReplIncompatible.
private void testTargetDbReplIncompatible(boolean setReplIncompProp) throws Throwable {
HiveConf primaryConf = primary.getConf();
TxnStore txnHandler = TxnUtils.getTxnStore(primary.getConf());
primary.run("use " + primaryDbName).run("CREATE TABLE t1(a string) STORED AS TEXTFILE").dump(primaryDbName);
replica.load(replicatedDbName, primaryDbName);
if (setReplIncompProp) {
replica.run("ALTER DATABASE " + replicatedDbName + " SET DBPROPERTIES('" + ReplConst.REPL_INCOMPATIBLE + "'='false')");
assert "false".equals(replica.getDatabase(replicatedDbName).getParameters().get(ReplConst.REPL_INCOMPATIBLE));
}
assertFalse(MetaStoreUtils.isDbReplIncompatible(replica.getDatabase(replicatedDbName)));
Long sourceTxnId = openTxns(1, txnHandler, primaryConf).get(0);
txnHandler.abortTxn(new AbortTxnRequest(sourceTxnId));
try {
sourceTxnId = openTxns(1, txnHandler, primaryConf).get(0);
primary.dump(primaryDbName);
replica.load(replicatedDbName, primaryDbName);
assertFalse(MetaStoreUtils.isDbReplIncompatible(replica.getDatabase(replicatedDbName)));
Long targetTxnId = txnHandler.getTargetTxnId(HiveUtils.getReplPolicy(replicatedDbName), sourceTxnId);
txnHandler.abortTxn(new AbortTxnRequest(targetTxnId));
assertTrue(MetaStoreUtils.isDbReplIncompatible(replica.getDatabase(replicatedDbName)));
WarehouseInstance.Tuple dumpData = primary.dump(primaryDbName);
assertFalse(ReplUtils.failedWithNonRecoverableError(new Path(dumpData.dumpLocation), conf));
replica.loadFailure(replicatedDbName, primaryDbName, null, ErrorMsg.REPL_INCOMPATIBLE_EXCEPTION.getErrorCode());
assertTrue(ReplUtils.failedWithNonRecoverableError(new Path(dumpData.dumpLocation), conf));
primary.dumpFailure(primaryDbName);
assertTrue(ReplUtils.failedWithNonRecoverableError(new Path(dumpData.dumpLocation), conf));
} finally {
txnHandler.abortTxn(new AbortTxnRequest(sourceTxnId));
}
}
use of org.apache.hadoop.hive.metastore.txn.TxnStore in project hive by apache.
the class TestReplicationScenariosAcidTables method testCompleteFailoverWithReverseBootstrap.
@Test
public void testCompleteFailoverWithReverseBootstrap() throws Throwable {
HiveConf primaryConf = primary.getConf();
TxnStore txnHandler = TxnUtils.getTxnStore(primary.getConf());
List<String> failoverConfigs = Arrays.asList("'" + HiveConf.ConfVars.HIVE_REPL_FAILOVER_START + "'='true'");
WarehouseInstance.Tuple dumpData = primary.run("use " + primaryDbName).run("create table t1 (id int) clustered by(id) into 3 buckets stored as orc " + "tblproperties (\"transactional\"=\"true\")").run("create table t2 (rank int) partitioned by (name string) tblproperties(\"transactional\"=\"true\", " + "\"transactional_properties\"=\"insert_only\")").dump(primaryDbName, failoverConfigs);
// This dump is not failover ready as target db can be used for replication only after first incremental load.
FileSystem fs = new Path(dumpData.dumpLocation).getFileSystem(conf);
Path dumpPath = new Path(dumpData.dumpLocation, ReplUtils.REPL_HIVE_BASE_DIR);
assertFalse(fs.exists(new Path(dumpPath, ReplAck.FAILOVER_READY_MARKER.toString())));
assertFalse(MetaStoreUtils.isDbBeingFailedOver(primary.getDatabase(primaryDbName)));
replica.load(replicatedDbName, primaryDbName, failoverConfigs).run("use " + replicatedDbName).run("show tables").verifyResults(new String[] { "t1", "t2" }).run("repl status " + replicatedDbName).verifyResult(dumpData.lastReplicationId);
Database db = replica.getDatabase(replicatedDbName);
assertTrue(MetaStoreUtils.isTargetOfReplication(db));
assertFalse(MetaStoreUtils.isDbBeingFailedOver(db));
primary.run("use " + primaryDbName).run("insert into t1 values(1)").run("insert into t2 partition(name='Bob') values(11)").run("insert into t2 partition(name='Carl') values(10)");
/**
*Open transactions can be of two types:
* Case 1) Txns that have not acquired HIVE LOCKS or they belong to different db: These txns would be captured in
* _failovermetadata file inside dump directory.
* Case 2) Txns that have acquired HIVE LOCKS and belong to db under replication: These txns would be aborted by hive
* as part of dump operation.
*/
// Open 3 txns for Database which is not under replication
int numTxnsForSecDb = 3;
List<Long> txnsForSecDb = openTxns(numTxnsForSecDb, txnHandler, primaryConf);
// Allocate write ids for both tables of secondary db for 3 txns
// t1=5 and t2=5
Map<String, Long> tablesInSecDb = new HashMap<>();
tablesInSecDb.put("t1", (long) numTxnsForSecDb);
tablesInSecDb.put("t2", (long) numTxnsForSecDb);
List<Long> lockIdsForSecDb = allocateWriteIdsForTablesAndAcquireLocks(primaryDbName + "_extra", tablesInSecDb, txnHandler, txnsForSecDb, primaryConf);
// Open 2 txns for Primary Db
int numTxnsForPrimaryDb = 2;
List<Long> txnsForPrimaryDb = openTxns(numTxnsForPrimaryDb, txnHandler, primaryConf);
// Allocate write ids for both tables of primary db for 2 txns
// t1=5 and t2=5
Map<String, Long> tablesInPrimaryDb = new HashMap<>();
tablesInPrimaryDb.put("t1", (long) numTxnsForPrimaryDb + 1);
tablesInPrimaryDb.put("t2", (long) numTxnsForPrimaryDb + 2);
List<Long> lockIdsForPrimaryDb = allocateWriteIdsForTablesAndAcquireLocks(primaryDbName, tablesInPrimaryDb, txnHandler, txnsForPrimaryDb, primaryConf);
// Open 1 txn with no hive locks acquired
List<Long> txnsWithNoLocks = openTxns(1, txnHandler, primaryConf);
dumpData = primary.dump(primaryDbName, failoverConfigs);
dumpPath = new Path(dumpData.dumpLocation, ReplUtils.REPL_HIVE_BASE_DIR);
assertTrue(fs.exists(new Path(dumpPath, DUMP_ACKNOWLEDGEMENT.toString())));
assertTrue(fs.exists(new Path(dumpPath, FailoverMetaData.FAILOVER_METADATA)));
assertTrue(fs.exists(new Path(dumpPath, ReplAck.FAILOVER_READY_MARKER.toString())));
assertTrue(MetaStoreUtils.isDbBeingFailedOverAtEndpoint(primary.getDatabase(primaryDbName), MetaStoreUtils.FailoverEndpoint.SOURCE));
FailoverMetaData failoverMD = new FailoverMetaData(dumpPath, conf);
List<Long> openTxns = failoverMD.getOpenTxns();
List<Long> txnsAborted = failoverMD.getAbortedTxns();
assertTrue(txnsAborted.size() == 2);
assertTrue(txnsAborted.containsAll(txnsForPrimaryDb));
assertTrue(openTxns.size() == 4);
assertTrue(openTxns.containsAll(txnsForSecDb));
assertTrue(openTxns.containsAll(txnsWithNoLocks));
assertTrue(failoverMD.getTxnsWithoutLock().equals(txnsWithNoLocks));
// TxnsForPrimaryDb and txnsWithNoLocks would have been aborted by dump operation.
verifyAllOpenTxnsAborted(txnsForPrimaryDb, primaryConf);
verifyAllOpenTxnsNotAborted(txnsForSecDb, primaryConf);
verifyAllOpenTxnsNotAborted(txnsWithNoLocks, primaryConf);
// Abort the txns
txnHandler.abortTxns(new AbortTxnsRequest(txnsForSecDb));
txnHandler.abortTxns(new AbortTxnsRequest(txnsWithNoLocks));
verifyAllOpenTxnsAborted(txnsForSecDb, primaryConf);
verifyAllOpenTxnsAborted(txnsWithNoLocks, primaryConf);
releaseLocks(txnHandler, lockIdsForSecDb);
replica.load(replicatedDbName, primaryDbName, failoverConfigs).run("use " + replicatedDbName).run("show tables").verifyResults(new String[] { "t1", "t2" }).run("repl status " + replicatedDbName).verifyResult(dumpData.lastReplicationId).run("select id from t1").verifyResults(new String[] { "1" }).run("select rank from t2 order by rank").verifyResults(new String[] { "10", "11" });
db = replica.getDatabase(replicatedDbName);
assertTrue(MetaStoreUtils.isTargetOfReplication(db));
assertTrue(MetaStoreUtils.isDbBeingFailedOverAtEndpoint(db, MetaStoreUtils.FailoverEndpoint.TARGET));
assertTrue(fs.exists(new Path(dumpPath, ReplAck.LOAD_ACKNOWLEDGEMENT.toString())));
Path dbRootDir = new Path(dumpData.dumpLocation).getParent();
long prevDumpDirModifTime = getLatestDumpDirModifTime(dbRootDir);
primary.run("REPL DUMP " + primaryDbName + " with ('" + HiveConf.ConfVars.HIVE_REPL_FAILOVER_START + "' = 'true')");
Assert.assertEquals(dumpData.dumpLocation, ReplUtils.getLatestDumpPath(dbRootDir, conf).toString());
Assert.assertEquals(prevDumpDirModifTime, getLatestDumpDirModifTime(dbRootDir));
dumpPath = new Path(dumpData.dumpLocation, ReplUtils.REPL_HIVE_BASE_DIR);
assertTrue(fs.exists(new Path(dumpPath, ReplAck.LOAD_ACKNOWLEDGEMENT.toString())));
assertTrue(MetaStoreUtils.isDbBeingFailedOverAtEndpoint(primary.getDatabase(primaryDbName), MetaStoreUtils.FailoverEndpoint.SOURCE));
primary.run("drop database if exists " + primaryDbName + " cascade");
assertTrue(primary.getDatabase(primaryDbName) == null);
assertFalse(ReplChangeManager.isSourceOfReplication(replica.getDatabase(replicatedDbName)));
WarehouseInstance.Tuple reverseDumpData = replica.run("create table t3 (id int)").run("insert into t2 partition(name='Bob') values(20)").run("insert into t3 values (2)").dump(replicatedDbName);
assertNotEquals(reverseDumpData.dumpLocation, dumpData.dumpLocation);
assertTrue(fs.exists(new Path(dumpPath, ReplAck.FAILOVER_READY_MARKER.toString())));
dumpPath = new Path(reverseDumpData.dumpLocation, ReplUtils.REPL_HIVE_BASE_DIR);
assertFalse(fs.exists(new Path(dumpPath, ReplAck.FAILOVER_READY_MARKER.toString())));
assertTrue(new DumpMetaData(dumpPath, conf).getDumpType() == DumpType.BOOTSTRAP);
assertTrue(fs.exists(new Path(dumpPath, DUMP_ACKNOWLEDGEMENT.toString())));
db = replica.getDatabase(replicatedDbName);
assertTrue(MetaStoreUtils.isDbBeingFailedOverAtEndpoint(db, MetaStoreUtils.FailoverEndpoint.TARGET));
assertFalse(MetaStoreUtils.isTargetOfReplication(db));
primary.load(primaryDbName, replicatedDbName).run("use " + primaryDbName).run("show tables").verifyResults(new String[] { "t1", "t2", "t3" }).run("repl status " + primaryDbName).verifyResult(reverseDumpData.lastReplicationId).run("select id from t1").verifyResults(new String[] { "1" }).run("select rank from t2 order by rank").verifyResults(new String[] { "10", "11", "20" }).run("select id from t3").verifyResults(new String[] { "2" });
Database primaryDb = primary.getDatabase(primaryDbName);
assertFalse(primaryDb == null);
assertTrue(ReplUtils.isFirstIncPending(primaryDb.getParameters()));
assertTrue(MetaStoreUtils.isTargetOfReplication(primaryDb));
assertFalse(MetaStoreUtils.isDbBeingFailedOver(primaryDb));
assertTrue(fs.exists(new Path(dumpPath, LOAD_ACKNOWLEDGEMENT.toString())));
assertFalse(ReplChangeManager.isSourceOfReplication(primaryDb));
assertTrue(ReplChangeManager.isSourceOfReplication(replica.getDatabase(replicatedDbName)));
reverseDumpData = replica.run("insert into t3 values (3)").run("insert into t2 partition(name='Bob') values(30)").dump(replicatedDbName);
assertFalse(MetaStoreUtils.isDbBeingFailedOver(replica.getDatabase(replicatedDbName)));
primary.load(primaryDbName, replicatedDbName).run("select rank from t2 order by rank").verifyResults(new String[] { "10", "11", "20", "30" }).run("select id from t3").verifyResults(new String[] { "2", "3" }).run("repl status " + primaryDbName).verifyResult(reverseDumpData.lastReplicationId);
assertFalse(ReplUtils.isFirstIncPending(primary.getDatabase(primaryDbName).getParameters()));
}
use of org.apache.hadoop.hive.metastore.txn.TxnStore in project hive by apache.
the class TestReplicationScenariosAcidTables method testAcidTablesBootstrapWithOpenTxnsTimeout.
@Test
public void testAcidTablesBootstrapWithOpenTxnsTimeout() throws Throwable {
int numTxns = 5;
HiveConf primaryConf = primary.getConf();
TxnStore txnHandler = TxnUtils.getTxnStore(primary.getConf());
// Open 5 txns
List<Long> txns = openTxns(numTxns, txnHandler, primaryConf);
// Create 2 tables, one partitioned and other not. Also, have both types of full ACID and MM tables.
primary.run("use " + primaryDbName).run("create table t1 (id int) clustered by(id) into 3 buckets stored as orc " + "tblproperties (\"transactional\"=\"true\")").run("insert into t1 values(1)").run("create table t2 (rank int) partitioned by (name string) tblproperties(\"transactional\"=\"true\", " + "\"transactional_properties\"=\"insert_only\")").run("insert into t2 partition(name='Bob') values(11)").run("insert into t2 partition(name='Carl') values(10)");
// Allocate write ids for both tables t1 and t2 for all txns
// t1=5+1(insert) and t2=5+2(insert)
Map<String, Long> tables = new HashMap<>();
tables.put("t1", numTxns + 1L);
tables.put("t2", numTxns + 2L);
List<Long> lockIds = allocateWriteIdsForTablesAndAcquireLocks(primaryDbName, tables, txnHandler, txns, primaryConf);
// Bootstrap dump with open txn timeout as 1s.
List<String> withConfigs = Arrays.asList("'" + HiveConf.ConfVars.REPL_BOOTSTRAP_DUMP_OPEN_TXN_TIMEOUT + "'='1s'");
WarehouseInstance.Tuple bootstrapDump = primary.run("use " + primaryDbName).dump(primaryDbName, withConfigs);
// After bootstrap dump, all the opened txns should be aborted. Verify it.
verifyAllOpenTxnsAborted(txns, primaryConf);
// Release the locks
releaseLocks(txnHandler, lockIds);
verifyNextId(tables, primaryDbName, primaryConf);
// Bootstrap load which should also replicate the aborted write ids on both tables.
HiveConf replicaConf = replica.getConf();
replica.load(replicatedDbName, primaryDbName).run("use " + replicatedDbName).run("show tables").verifyResults(new String[] { "t1", "t2" }).run("repl status " + replicatedDbName).verifyResult(bootstrapDump.lastReplicationId).run("select id from t1").verifyResults(new String[] { "1" }).run("select rank from t2 order by rank").verifyResults(new String[] { "10", "11" });
// Verify if HWM is properly set after REPL LOAD
verifyNextId(tables, replicatedDbName, replicaConf);
// Verify if all the aborted write ids are replicated to the replicated DB
for (Map.Entry<String, Long> entry : tables.entrySet()) {
entry.setValue((long) numTxns);
}
verifyWriteIdsForTables(tables, replicaConf, replicatedDbName);
// Verify if entries added in COMPACTION_QUEUE for each table/partition
// t1-> 1 entry and t2-> 2 entries (1 per partition)
tables.clear();
tables.put("t1", 1L);
tables.put("t2", 2L);
verifyCompactionQueue(tables, replicatedDbName, replicaConf);
}
use of org.apache.hadoop.hive.metastore.txn.TxnStore in project hive by apache.
the class TestReplicationScenariosAcidTablesBootstrap method testAcidTablesBootstrapDuringIncrementalWithOpenTxnsTimeout.
@Test
public void testAcidTablesBootstrapDuringIncrementalWithOpenTxnsTimeout() throws Throwable {
// Take a dump without ACID tables
WarehouseInstance.Tuple bootstrapDump = prepareDataAndDump(primaryDbName, dumpWithoutAcidClause);
LOG.info(testName.getMethodName() + ": loading dump without acid tables.");
replica.load(replicatedDbName, primaryDbName);
// Open concurrent transactions, create data for incremental and take an incremental dump
// with ACID table bootstrap.
int numTxns = 5;
HiveConf primaryConf = primary.getConf();
TxnStore txnHandler = TxnUtils.getTxnStore(primary.getConf());
// Open 5 txns
List<Long> txns = openTxns(numTxns, txnHandler, primaryConf);
prepareIncNonAcidData(primaryDbName);
prepareIncAcidData(primaryDbName);
// Allocate write ids for tables t1 and t2 for all txns
// t1=5+2(insert) and t2=5+6(insert, alter add column), now alter also creates a transaction
Map<String, Long> tables = new HashMap<>();
tables.put("t1", numTxns + 2L);
tables.put("t2", numTxns + 6L);
List<Long> lockIds = allocateWriteIdsForTablesAndAcquireLocks(primaryDbName, tables, txnHandler, txns, primaryConf);
// Bootstrap dump with open txn timeout as 1s.
List<String> withConfigs = new LinkedList<>(dumpWithAcidBootstrapClause);
withConfigs.add("'hive.repl.bootstrap.dump.open.txn.timeout'='1s'");
WarehouseInstance.Tuple incDump = primary.run("use " + primaryDbName).dump(primaryDbName, withConfigs);
// After bootstrap dump, all the opened txns should be aborted. Verify it.
verifyAllOpenTxnsAborted(txns, primaryConf);
releaseLocks(txnHandler, lockIds);
verifyNextId(tables, primaryDbName, primaryConf);
// Incremental load with ACID bootstrap should also replicate the aborted write ids on
// tables t1 and t2
HiveConf replicaConf = replica.getConf();
LOG.info(testName.getMethodName() + ": loading incremental dump with ACID bootstrap.");
replica.load(replicatedDbName, primaryDbName);
verifyIncLoad(replicatedDbName, incDump.lastReplicationId);
// Verify if HWM is properly set after REPL LOAD
verifyNextId(tables, replicatedDbName, replicaConf);
// Verify if all the aborted write ids are replicated to the replicated DB
for (Map.Entry<String, Long> entry : tables.entrySet()) {
entry.setValue((long) numTxns);
}
verifyWriteIdsForTables(tables, replicaConf, replicatedDbName);
// Verify if entries added in COMPACTION_QUEUE for each table/partition
// t1-> 1 entry and t2-> 2 entries (1 per partition)
tables.clear();
tables.put("t1", 1L);
tables.put("t2", 4L);
verifyCompactionQueue(tables, replicatedDbName, replicaConf);
}
use of org.apache.hadoop.hive.metastore.txn.TxnStore in project hive by apache.
the class TestCrudCompactorOnTez method testMinorCompactionShouldBeRefusedOnTablesWithOriginalFiles.
@Test
public void testMinorCompactionShouldBeRefusedOnTablesWithOriginalFiles() throws Exception {
conf.setBoolVar(HiveConf.ConfVars.COMPACTOR_CRUD_QUERY_BASED, true);
// Set delta numbuer threshold to 2 to avoid skipping compaction because of too few deltas
conf.setIntVar(HiveConf.ConfVars.HIVE_COMPACTOR_DELTA_NUM_THRESHOLD, 2);
// Set delta percentage to a high value to suppress selecting major compression based on that
conf.setFloatVar(HiveConf.ConfVars.HIVE_COMPACTOR_DELTA_PCT_THRESHOLD, 1000f);
final String dbName = "default";
final String tableName = "compaction_test";
executeStatementOnDriver("drop table if exists " + tableName, driver);
executeStatementOnDriver("CREATE TABLE " + tableName + "(id string, value string) CLUSTERED BY(id) " + "INTO 10 BUCKETS STORED AS ORC TBLPROPERTIES('transactional'='false')", driver);
executeStatementOnDriver("INSERT INTO TABLE " + tableName + " values ('1','one'),('2','two'),('3','three')," + "('4','four'),('5','five'),('6','six'),('7','seven'),('8','eight'),('9','nine'),('10','ten')," + "('11','eleven'),('12','twelve'),('13','thirteen'),('14','fourteen'),('15','fifteen'),('16','sixteen')," + "('17','seventeen'),('18','eighteen'),('19','nineteen'),('20','twenty')", driver);
executeStatementOnDriver("alter table " + tableName + " set TBLPROPERTIES('transactional'='true')", driver);
executeStatementOnDriver("insert into " + tableName + " values ('21', 'value21'),('84', 'value84')," + "('66', 'value66'),('54', 'value54')", driver);
executeStatementOnDriver("insert into " + tableName + " values ('22', 'value22'),('34', 'value34')," + "('35', 'value35')", driver);
executeStatementOnDriver("insert into " + tableName + " values ('75', 'value75'),('99', 'value99')", driver);
execSelectAndDumpData("select * from " + tableName, driver, "Dumping data for " + tableName + " after load:");
TxnStore txnHandler = TxnUtils.getTxnStore(conf);
// Prevent initiator from submitting the compaction requests
TxnStore mockedHandler = spy(txnHandler);
doThrow(new RuntimeException("")).when(mockedHandler).compact(nullable(CompactionRequest.class));
Initiator initiator = new Initiator();
initiator.setConf(conf);
initiator.init(new AtomicBoolean(true));
FieldSetter.setField(initiator, MetaStoreCompactorThread.class.getDeclaredField("txnHandler"), mockedHandler);
// Run initiator and capture compaction requests
initiator.run();
// Check captured compaction request and if the type for the table was MAJOR
ArgumentCaptor<CompactionRequest> requests = ArgumentCaptor.forClass(CompactionRequest.class);
verify(mockedHandler).compact(requests.capture());
Assert.assertTrue(requests.getAllValues().stream().anyMatch(r -> r.getTablename().equals(tableName) && r.getType().equals(CompactionType.MAJOR)));
// Try to do a minor compaction directly
CompactionRequest rqst = new CompactionRequest(dbName, tableName, CompactionType.MINOR);
txnHandler.compact(rqst);
runWorker(conf);
// Check if both compactions were failed with the expected error message
ShowCompactResponse rsp = txnHandler.showCompact(new ShowCompactRequest());
List<ShowCompactResponseElement> compacts = rsp.getCompacts();
if (2 != compacts.size()) {
Assert.fail("Expecting 2 rows and found " + compacts.size() + " files " + compacts);
}
Assert.assertEquals("did not initiate", compacts.get(0).getState());
Assert.assertTrue(compacts.get(0).getErrorMessage().startsWith("Caught exception while trying to determine if we should compact"));
Assert.assertEquals("refused", compacts.get(1).getState());
Assert.assertTrue(compacts.get(1).getErrorMessage().startsWith("Query based Minor compaction is not possible for full acid tables having raw format (non-acid) data in them."));
}
Aggregations