From 0e9ef913633f9e403e6b944f5e738066ed9be386 Mon Sep 17 00:00:00 2001 From: Pankaj Kumar Date: Tue, 20 Jun 2017 21:01:42 +0800 Subject: [PATCH] HBASE-18167 OfflineMetaRepair tool may cause HMaster abort always --- .../org/apache/hadoop/hbase/MetaTableAccessor.java | 35 +++++++++++ .../hadoop/hbase/master/AssignmentManager.java | 66 ++++++++++++++++++--- .../util/hbck/TestOfflineMetaRebuildBase.java | 69 +++++++++++++++++++++- 3 files changed, 159 insertions(+), 11 deletions(-) diff --git a/hbase-client/src/main/java/org/apache/hadoop/hbase/MetaTableAccessor.java b/hbase-client/src/main/java/org/apache/hadoop/hbase/MetaTableAccessor.java index 2bbae15..9d32923 100644 --- a/hbase-client/src/main/java/org/apache/hadoop/hbase/MetaTableAccessor.java +++ b/hbase-client/src/main/java/org/apache/hadoop/hbase/MetaTableAccessor.java @@ -1721,4 +1721,39 @@ public class MetaTableAccessor { return null; } + /** + * Checks whether hbase:meta contains any info:server entry. + * @param connection connection we're using + * @return true if hbase:meta contains any info:server entry, false if not + * @throws IOException + */ + public static boolean infoServerExists(Connection connection) throws IOException { + // Make a version of ResultCollectingVisitor that only collects the first + CollectingVisitor visitor = new CollectingVisitor() { + @Override + public boolean visit(Result r) throws IOException { + if (r == null || r.isEmpty()) return true; + RegionLocations locations = getRegionLocations(r); + if (locations == null) return true; + for (HRegionLocation loc : locations.getRegionLocations()) { + if (loc != null) { + if (loc.getServerName() != null) { + add(r); + // Stop collecting results after we get one. + return false; + } + } + } + return true; + } + + @Override + void add(Result r) { + this.results.add(r); + } + }; + fullScan(connection, visitor); + // If visitor has results >= 1 then hbase:meta has the info:server entry + return visitor.getResults().size() >= 1; + } } diff --git a/hbase-server/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java b/hbase-server/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java index 0a28967..cb85ca2 100644 --- a/hbase-server/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java +++ b/hbase-server/src/main/java/org/apache/hadoop/hbase/master/AssignmentManager.java @@ -634,17 +634,10 @@ public class AssignmentManager extends ZooKeeperListener { } } - Set disabledOrDisablingOrEnabling = null; Map allRegions = null; - if (!failover) { - disabledOrDisablingOrEnabling = tableStateManager.getTablesInStates( - ZooKeeperProtos.Table.State.DISABLED, ZooKeeperProtos.Table.State.DISABLING, - ZooKeeperProtos.Table.State.ENABLING); - - // Clean re/start, mark all user regions closed before reassignment - allRegions = regionStates.closeAllUserRegions( - disabledOrDisablingOrEnabling); + // Retrieve user regions except tables region that are in disabled/disabling/enabling states. + allRegions = getUserRegionsToAssign(); } // Now region states are restored @@ -656,6 +649,15 @@ public class AssignmentManager extends ZooKeeperListener { // Process list of dead servers and regions in RIT. // See HBASE-4580 for more information. processDeadServersAndRecoverLostRegions(deadServers); + + // Handle the scenario when meta is rebuild by OfflineMetaRepair tool. + // In this scenario, meta will have only info:regioninfo entries (won't contain info:server) + // which lead SSH/SCP to skip holding region assignment. + if (!MetaTableAccessor.infoServerExists(server.getConnection())) { + // Need to assign the user region as a fresh startup, otherwise user region assignment will + // never happen + assignRegionsOnSSHCompletion(); + } } if (!failover && useZKForAssignment) { @@ -685,6 +687,52 @@ public class AssignmentManager extends ZooKeeperListener { return failover; } + /* + * At cluster clean re/start, mark all user regions closed except those of tables that are + * excluded, such as disabled/disabling/enabling tables. All user regions and their previous + * locations are returned. + */ + private Map getUserRegionsToAssign() + throws InterruptedIOException, CoordinatedStateException { + Set disabledOrDisablingOrEnabling = + tableStateManager.getTablesInStates(ZooKeeperProtos.Table.State.DISABLED, + ZooKeeperProtos.Table.State.DISABLING, ZooKeeperProtos.Table.State.ENABLING); + + // Clean re/start, mark all user regions closed before reassignment + return regionStates.closeAllUserRegions(disabledOrDisablingOrEnabling); + } + + /* + * Wait for SSH completion and assign user region which are not in disabled/disabling/enabling + * table states. + */ + private void assignRegionsOnSSHCompletion() { + LOG.info("Meta is rebuild by OfflineMetaRepair tool, assigning all user regions."); + Thread regionAssignerThread = new Thread("RegionAssignerOnMetaRebuild") { + public void run() { + // Wait until all dead server processing finish + while (serverManager.areDeadServersInProgress()) { + try { + Thread.sleep(100); + } catch (InterruptedException e) { + LOG.warn("RegionAssignerOnMetaRebuild got interrupted.", e); + Thread.currentThread().interrupt(); + return; + } + } + LOG.info("SSH has been completed for all dead servers, assigning user regions."); + try { + // Assign the regions + assignAllUserRegions(getUserRegionsToAssign()); + } catch (CoordinatedStateException | IOException | InterruptedException e) { + LOG.error("Exception occured while assigning user regions.", e); + } + }; + }; + regionAssignerThread.setDaemon(true); + regionAssignerThread.start(); + } + /** * If region is up in zk in transition, then do fixup and block and wait until * the region is assigned and out of transition. Used on startup for diff --git a/hbase-server/src/test/java/org/apache/hadoop/hbase/util/hbck/TestOfflineMetaRebuildBase.java b/hbase-server/src/test/java/org/apache/hadoop/hbase/util/hbck/TestOfflineMetaRebuildBase.java index b31e20e..e764f3e 100644 --- a/hbase-server/src/test/java/org/apache/hadoop/hbase/util/hbck/TestOfflineMetaRebuildBase.java +++ b/hbase-server/src/test/java/org/apache/hadoop/hbase/util/hbck/TestOfflineMetaRebuildBase.java @@ -25,6 +25,7 @@ import static org.junit.Assert.assertTrue; import java.io.IOException; import java.util.Arrays; +import java.util.List; import org.apache.commons.logging.Log; import org.apache.commons.logging.LogFactory; @@ -33,13 +34,17 @@ import org.apache.hadoop.fs.FileSystem; import org.apache.hadoop.fs.Path; import org.apache.hadoop.hbase.HConstants; import org.apache.hadoop.hbase.HTableDescriptor; +import org.apache.hadoop.hbase.NamespaceDescriptor; +import org.apache.hadoop.hbase.TableName; import org.apache.hadoop.hbase.testclassification.MediumTests; import org.apache.hadoop.hbase.client.Admin; import org.apache.hadoop.hbase.client.Connection; import org.apache.hadoop.hbase.client.ConnectionFactory; +import org.apache.hadoop.hbase.util.Bytes; import org.apache.hadoop.hbase.util.FSUtils; import org.apache.hadoop.hbase.util.HBaseFsck; import org.apache.hadoop.hbase.util.HBaseFsck.ErrorReporter.ERROR_CODE; +import org.apache.hadoop.hbase.util.JVMClusterUtil.RegionServerThread; import org.junit.Test; import org.junit.experimental.categories.Category; @@ -77,6 +82,66 @@ public class TestOfflineMetaRebuildBase extends OfflineMetaRebuildTestCore { // bring up the minicluster TEST_UTIL.startMiniZKCluster(); TEST_UTIL.restartHBaseCluster(3); + validateMetaAndUserTableRows(1, 5); + } + + @Test(timeout = 300000) + public void testHMasterStartupOnMetaRebuild() throws Exception { + // shutdown the minicluster + TEST_UTIL.shutdownMiniHBaseCluster(); + + // Assign meta in master and restart HBase cluster + TEST_UTIL.getConfiguration().set("hbase.balancer.tablesOnMaster", "hbase:meta"); + // Set namespace initialization timeout + TEST_UTIL.getConfiguration().set("hbase.master.namespace.init.timeout", "150000"); + TEST_UTIL.restartHBaseCluster(3); + TEST_UTIL.getMiniHBaseCluster().waitForActiveAndReadyMaster(); + + try { + // Create namespace + TEST_UTIL.getHBaseAdmin().createNamespace(NamespaceDescriptor.create("ns1").build()); + TEST_UTIL.getHBaseAdmin().createNamespace(NamespaceDescriptor.create("ns2").build()); + // Create tables + TEST_UTIL.createTable(TableName.valueOf("ns1:testHMasterStartupOnMetaRebuild"), + Bytes.toBytes("cf1")); + TEST_UTIL.createTable(TableName.valueOf("ns2:testHMasterStartupOnMetaRebuild"), + Bytes.toBytes("cf1")); + + // Flush meta + TEST_UTIL.flush(TableName.META_TABLE_NAME); + + // HMaster graceful shutdown + TEST_UTIL.getHBaseCluster().getMaster().shutdown(); + + // Kill region servers + List regionServerThreads = + TEST_UTIL.getHBaseCluster().getRegionServerThreads(); + for (RegionServerThread regionServerThread : regionServerThreads) { + TEST_UTIL.getHBaseCluster() + .killRegionServer(regionServerThread.getRegionServer().getServerName()); + } + + // rebuild meta table from scratch + HBaseFsck fsck = new HBaseFsck(conf); + assertTrue(fsck.rebuildMeta(false)); + + // bring up the minicluster + TEST_UTIL.restartHBaseCluster(3); + validateMetaAndUserTableRows(3, 7); + } finally { + // Remove table and namesapce + TEST_UTIL.deleteTable("ns1:testHMasterStartupOnMetaRebuild"); + TEST_UTIL.deleteTable("ns2:testHMasterStartupOnMetaRebuild"); + TEST_UTIL.getHBaseAdmin().deleteNamespace("ns1"); + TEST_UTIL.getHBaseAdmin().deleteNamespace("ns2"); + } + } + + /* + * Validate meta table region count and user table rows. + */ + private void validateMetaAndUserTableRows(int totalTableCount, int totalRegionCount) + throws Exception { try (Connection connection = ConnectionFactory.createConnection(TEST_UTIL.getConfiguration())) { Admin admin = connection.getAdmin(); admin.enableTable(table); @@ -85,10 +150,10 @@ public class TestOfflineMetaRebuildBase extends OfflineMetaRebuildTestCore { LOG.info("No more RIT in ZK, now doing final test verification"); // everything is good again. - assertEquals(5, scanMeta()); + assertEquals(totalRegionCount, scanMeta()); HTableDescriptor[] htbls = admin.listTables(); LOG.info("Tables present after restart: " + Arrays.toString(htbls)); - assertEquals(1, htbls.length); + assertEquals(totalTableCount, htbls.length); } assertErrors(doFsck(conf, false), new ERROR_CODE[] {}); -- 2.7.2.windows.1