org.apache.hadoop.hbase.master.procedure.ServerCrashProcedure Maven / Gradle / Ivy
Show all versions of hbase-server Show documentation
/*
* 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.hbase.master.procedure;
import static org.apache.hadoop.hbase.HConstants.DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK;
import static org.apache.hadoop.hbase.HConstants.HBASE_SPLIT_WAL_COORDINATED_BY_ZK;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import org.apache.hadoop.hbase.DoNotRetryIOException;
import org.apache.hadoop.hbase.ServerName;
import org.apache.hadoop.hbase.client.RegionInfo;
import org.apache.hadoop.hbase.client.RegionInfoBuilder;
import org.apache.hadoop.hbase.client.RegionReplicaUtil;
import org.apache.hadoop.hbase.client.TableState;
import org.apache.hadoop.hbase.master.MasterServices;
import org.apache.hadoop.hbase.master.MasterWalManager;
import org.apache.hadoop.hbase.master.SplitWALManager;
import org.apache.hadoop.hbase.master.assignment.AssignmentManager;
import org.apache.hadoop.hbase.master.assignment.RegionStateNode;
import org.apache.hadoop.hbase.master.assignment.TransitRegionStateProcedure;
import org.apache.hadoop.hbase.monitoring.MonitoredTask;
import org.apache.hadoop.hbase.monitoring.TaskMonitor;
import org.apache.hadoop.hbase.procedure2.Procedure;
import org.apache.hadoop.hbase.procedure2.ProcedureMetrics;
import org.apache.hadoop.hbase.procedure2.ProcedureStateSerializer;
import org.apache.hadoop.hbase.procedure2.ProcedureSuspendedException;
import org.apache.hadoop.hbase.procedure2.ProcedureYieldException;
import org.apache.hadoop.hbase.procedure2.StateMachineProcedure;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos;
import org.apache.hadoop.hbase.shaded.protobuf.generated.MasterProcedureProtos.ServerCrashState;
/**
* Handle crashed server. This is a port to ProcedureV2 of what used to be euphemistically called
* ServerShutdownHandler.
*
* The procedure flow varies dependent on whether meta is assigned and if we are to split logs.
*
* We come in here after ServerManager has noticed a server has expired. Procedures queued on the
* rpc should have been notified about fail and should be concurrently getting themselves ready to
* assign elsewhere.
*/
@InterfaceAudience.Private
public class ServerCrashProcedure extends
StateMachineProcedure implements ServerProcedureInterface {
private static final Logger LOG = LoggerFactory.getLogger(ServerCrashProcedure.class);
/**
* Configuration parameter to enable/disable the retain region assignment during
* ServerCrashProcedure.
*
* By default retain assignment is disabled which makes the failover faster and improve the
* availability; useful for cloud scenario where region block locality is not important. Enable
* this when RegionServers are deployed on same host where Datanode are running, this will improve
* read performance due to local read.
*
* see HBASE-24900 for more details.
*/
public static final String MASTER_SCP_RETAIN_ASSIGNMENT = "hbase.master.scp.retain.assignment";
/** Default value of {@link #MASTER_SCP_RETAIN_ASSIGNMENT} */
public static final boolean DEFAULT_MASTER_SCP_RETAIN_ASSIGNMENT = false;
/**
* Name of the crashed server to process.
*/
private ServerName serverName;
/**
* Whether DeadServer knows that we are processing it.
*/
private boolean notifiedDeadServer = false;
/**
* Regions that were on the crashed server.
*/
private List regionsOnCrashedServer;
private boolean carryingMeta = false;
private boolean shouldSplitWal;
private MonitoredTask status;
// currentRunningState is updated when ServerCrashProcedure get scheduled, child procedures update
// progress will not update the state because the actual state is overwritten by its next state
private ServerCrashState currentRunningState = getInitialState();
/**
* Call this constructor queuing up a Procedure.
* @param serverName Name of the crashed server.
* @param shouldSplitWal True if we should split WALs as part of crashed server processing.
* @param carryingMeta True if carrying hbase:meta table region.
*/
public ServerCrashProcedure(final MasterProcedureEnv env, final ServerName serverName,
final boolean shouldSplitWal, final boolean carryingMeta) {
this.serverName = serverName;
this.shouldSplitWal = shouldSplitWal;
this.carryingMeta = carryingMeta;
this.setOwner(env.getRequestUser());
}
/**
* Used when deserializing from a procedure store; we'll construct one of these then call
* #deserializeStateData(InputStream). Do not use directly.
*/
public ServerCrashProcedure() {
}
public boolean isInRecoverMetaState() {
return getCurrentState() == ServerCrashState.SERVER_CRASH_PROCESS_META;
}
@Override
protected Flow executeFromState(MasterProcedureEnv env, ServerCrashState state)
throws ProcedureSuspendedException, ProcedureYieldException {
final MasterServices services = env.getMasterServices();
final AssignmentManager am = env.getAssignmentManager();
updateProgress(true);
// HBASE-14802 If we have not yet notified that we are processing a dead server, do so now.
// This adds server to the DeadServer processing list but not to the DeadServers list.
// Server gets removed from processing list below on procedure successful finish.
if (!notifiedDeadServer) {
services.getServerManager().getDeadServers().processing(serverName);
notifiedDeadServer = true;
}
switch (state) {
case SERVER_CRASH_START:
case SERVER_CRASH_SPLIT_META_LOGS:
case SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR:
case SERVER_CRASH_ASSIGN_META:
break;
default:
// If hbase:meta is not assigned, yield.
if (env.getAssignmentManager().waitMetaLoaded(this)) {
throw new ProcedureSuspendedException();
}
}
try {
switch (state) {
case SERVER_CRASH_START:
LOG.info("Start " + this);
// If carrying meta, process it first. Else, get list of regions on crashed server.
if (this.carryingMeta) {
setNextState(ServerCrashState.SERVER_CRASH_SPLIT_META_LOGS);
} else {
setNextState(ServerCrashState.SERVER_CRASH_GET_REGIONS);
}
break;
case SERVER_CRASH_SPLIT_META_LOGS:
if (
env.getMasterConfiguration().getBoolean(HBASE_SPLIT_WAL_COORDINATED_BY_ZK,
DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK)
) {
zkCoordinatedSplitMetaLogs(env);
setNextState(ServerCrashState.SERVER_CRASH_ASSIGN_META);
} else {
am.getRegionStates().metaLogSplitting(serverName);
addChildProcedure(createSplittingWalProcedures(env, true));
setNextState(ServerCrashState.SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR);
}
break;
case SERVER_CRASH_DELETE_SPLIT_META_WALS_DIR:
if (isSplittingDone(env, true)) {
setNextState(ServerCrashState.SERVER_CRASH_ASSIGN_META);
am.getRegionStates().metaLogSplit(serverName);
} else {
setNextState(ServerCrashState.SERVER_CRASH_SPLIT_META_LOGS);
}
break;
case SERVER_CRASH_ASSIGN_META:
assignRegions(env, Arrays.asList(RegionInfoBuilder.FIRST_META_REGIONINFO));
setNextState(ServerCrashState.SERVER_CRASH_GET_REGIONS);
break;
case SERVER_CRASH_GET_REGIONS:
this.regionsOnCrashedServer = getRegionsOnCrashedServer(env);
// Where to go next? Depends on whether we should split logs at all or
// if we should do distributed log splitting.
if (regionsOnCrashedServer != null) {
LOG.info("{} had {} regions", serverName, regionsOnCrashedServer.size());
if (LOG.isTraceEnabled()) {
this.regionsOnCrashedServer.stream().forEach(ri -> LOG.trace(ri.getShortNameToLog()));
}
}
if (!this.shouldSplitWal) {
setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);
} else {
setNextState(ServerCrashState.SERVER_CRASH_SPLIT_LOGS);
}
break;
case SERVER_CRASH_SPLIT_LOGS:
if (
env.getMasterConfiguration().getBoolean(HBASE_SPLIT_WAL_COORDINATED_BY_ZK,
DEFAULT_HBASE_SPLIT_COORDINATED_BY_ZK)
) {
zkCoordinatedSplitLogs(env);
setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);
} else {
am.getRegionStates().logSplitting(this.serverName);
addChildProcedure(createSplittingWalProcedures(env, false));
setNextState(ServerCrashState.SERVER_CRASH_DELETE_SPLIT_WALS_DIR);
}
break;
case SERVER_CRASH_DELETE_SPLIT_WALS_DIR:
if (isSplittingDone(env, false)) {
cleanupSplitDir(env);
setNextState(ServerCrashState.SERVER_CRASH_ASSIGN);
am.getRegionStates().logSplit(this.serverName);
} else {
setNextState(ServerCrashState.SERVER_CRASH_SPLIT_LOGS);
}
break;
case SERVER_CRASH_ASSIGN:
// If no regions to assign, skip assign and skip to the finish.
// Filter out meta regions. Those are handled elsewhere in this procedure.
// Filter changes this.regionsOnCrashedServer.
if (filterDefaultMetaRegions()) {
if (LOG.isTraceEnabled()) {
LOG.trace("Assigning regions " + RegionInfo.getShortNameToLog(regionsOnCrashedServer)
+ ", " + this + "; cycles=" + getCycles());
}
assignRegions(env, regionsOnCrashedServer);
}
setNextState(ServerCrashState.SERVER_CRASH_FINISH);
break;
case SERVER_CRASH_HANDLE_RIT2:
// Noop. Left in place because we used to call handleRIT here for a second time
// but no longer necessary since HBASE-20634.
setNextState(ServerCrashState.SERVER_CRASH_FINISH);
break;
case SERVER_CRASH_FINISH:
LOG.info("removed crashed server {} after splitting done", serverName);
services.getAssignmentManager().getRegionStates().removeServer(serverName);
services.getServerManager().getDeadServers().finish(serverName);
updateProgress(true);
return Flow.NO_MORE_STATE;
default:
throw new UnsupportedOperationException("unhandled state=" + state);
}
} catch (IOException e) {
LOG.warn("Failed state=" + state + ", retry " + this + "; cycles=" + getCycles(), e);
}
return Flow.HAS_MORE_STATE;
}
/** Returns List of Regions on crashed server. */
List getRegionsOnCrashedServer(MasterProcedureEnv env) {
return env.getMasterServices().getAssignmentManager().getRegionsOnServer(serverName);
}
private void cleanupSplitDir(MasterProcedureEnv env) {
SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();
try {
if (!this.carryingMeta) {
// If we are NOT carrying hbase:meta, check if any left-over hbase:meta WAL files from an
// old hbase:meta tenancy on this server; clean these up if any before trying to remove the
// WAL directory of this server or we will fail. See archiveMetaLog comment for more details
// on this condition.
env.getMasterServices().getMasterWalManager().archiveMetaLog(this.serverName);
}
splitWALManager.deleteWALDir(serverName);
} catch (IOException e) {
LOG.info("Remove WAL directory for {} failed, ignore...{}", serverName, e.getMessage());
}
}
private boolean isSplittingDone(MasterProcedureEnv env, boolean splitMeta) {
SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();
try {
int wals = splitWALManager.getWALsToSplit(serverName, splitMeta).size();
LOG.debug("Check if {} WAL splitting is done? wals={}, meta={}", serverName, wals, splitMeta);
return wals == 0;
} catch (IOException e) {
LOG.warn("Get WALs of {} failed, retry...", serverName, e);
return false;
}
}
private Procedure[] createSplittingWalProcedures(MasterProcedureEnv env, boolean splitMeta)
throws IOException {
LOG.info("Splitting WALs {}, isMeta: {}", this, splitMeta);
SplitWALManager splitWALManager = env.getMasterServices().getSplitWALManager();
List procedures = splitWALManager.splitWALs(serverName, splitMeta);
return procedures.toArray(new Procedure[procedures.size()]);
}
private boolean filterDefaultMetaRegions() {
if (regionsOnCrashedServer == null) {
return false;
}
regionsOnCrashedServer.removeIf(this::isDefaultMetaRegion);
return !regionsOnCrashedServer.isEmpty();
}
private boolean isDefaultMetaRegion(RegionInfo hri) {
return hri.isMetaRegion() && RegionReplicaUtil.isDefaultReplica(hri);
}
/**
* Split hbase:meta logs using 'classic' zk-based coordination. Superceded by procedure-based WAL
* splitting.
* @see #createSplittingWalProcedures(MasterProcedureEnv, boolean)
*/
private void zkCoordinatedSplitMetaLogs(MasterProcedureEnv env) throws IOException {
LOG.debug("Splitting meta WALs {}", this);
MasterWalManager mwm = env.getMasterServices().getMasterWalManager();
AssignmentManager am = env.getMasterServices().getAssignmentManager();
am.getRegionStates().metaLogSplitting(serverName);
mwm.splitMetaLog(serverName);
am.getRegionStates().metaLogSplit(serverName);
LOG.debug("Done splitting meta WALs {}", this);
}
/**
* Split logs using 'classic' zk-based coordination. Superceded by procedure-based WAL splitting.
* @see #createSplittingWalProcedures(MasterProcedureEnv, boolean)
*/
private void zkCoordinatedSplitLogs(final MasterProcedureEnv env) throws IOException {
LOG.debug("Splitting WALs {}", this);
MasterWalManager mwm = env.getMasterServices().getMasterWalManager();
AssignmentManager am = env.getMasterServices().getAssignmentManager();
// TODO: For Matteo. Below BLOCKs!!!! Redo so can relinquish executor while it is running.
// PROBLEM!!! WE BLOCK HERE. Can block for hours if hundreds of WALs to split and hundreds
// of SCPs running because big cluster crashed down.
am.getRegionStates().logSplitting(this.serverName);
mwm.splitLog(this.serverName);
if (!carryingMeta) {
mwm.archiveMetaLog(this.serverName);
}
am.getRegionStates().logSplit(this.serverName);
LOG.debug("Done splitting WALs {}", this);
}
void updateProgress(boolean updateState) {
String msg = "Processing ServerCrashProcedure of " + serverName;
if (status == null) {
status = TaskMonitor.get().createStatus(msg);
return;
}
if (currentRunningState == ServerCrashState.SERVER_CRASH_FINISH) {
status.markComplete(msg + " done");
return;
}
if (updateState) {
currentRunningState = getCurrentState();
}
int childrenLatch = getChildrenLatch();
status.setStatus(msg + " current State " + currentRunningState
+ (childrenLatch > 0
? "; remaining num of running child procedures = " + childrenLatch
: ""));
}
@Override
protected void rollbackState(MasterProcedureEnv env, ServerCrashState state) throws IOException {
// Can't rollback.
throw new UnsupportedOperationException("unhandled state=" + state);
}
@Override
protected ServerCrashState getState(int stateId) {
return ServerCrashState.forNumber(stateId);
}
@Override
protected int getStateId(ServerCrashState state) {
return state.getNumber();
}
@Override
protected ServerCrashState getInitialState() {
return ServerCrashState.SERVER_CRASH_START;
}
@Override
protected boolean abort(MasterProcedureEnv env) {
// TODO
return false;
}
@Override
protected LockState acquireLock(final MasterProcedureEnv env) {
if (env.getProcedureScheduler().waitServerExclusiveLock(this, getServerName())) {
return LockState.LOCK_EVENT_WAIT;
}
return LockState.LOCK_ACQUIRED;
}
@Override
protected void releaseLock(final MasterProcedureEnv env) {
env.getProcedureScheduler().wakeServerExclusiveLock(this, getServerName());
}
@Override
public void toStringClassDetails(StringBuilder sb) {
sb.append(getProcName());
sb.append(", splitWal=");
sb.append(shouldSplitWal);
sb.append(", meta=");
sb.append(carryingMeta);
}
@Override
public String getProcName() {
return getClass().getSimpleName() + " " + this.serverName;
}
@Override
protected void serializeStateData(ProcedureStateSerializer serializer) throws IOException {
super.serializeStateData(serializer);
MasterProcedureProtos.ServerCrashStateData.Builder state =
MasterProcedureProtos.ServerCrashStateData.newBuilder()
.setServerName(ProtobufUtil.toServerName(this.serverName))
.setCarryingMeta(this.carryingMeta).setShouldSplitWal(this.shouldSplitWal);
if (this.regionsOnCrashedServer != null && !this.regionsOnCrashedServer.isEmpty()) {
for (RegionInfo hri : this.regionsOnCrashedServer) {
state.addRegionsOnCrashedServer(ProtobufUtil.toRegionInfo(hri));
}
}
serializer.serialize(state.build());
}
@Override
protected void deserializeStateData(ProcedureStateSerializer serializer) throws IOException {
super.deserializeStateData(serializer);
MasterProcedureProtos.ServerCrashStateData state =
serializer.deserialize(MasterProcedureProtos.ServerCrashStateData.class);
this.serverName = ProtobufUtil.toServerName(state.getServerName());
this.carryingMeta = state.hasCarryingMeta() ? state.getCarryingMeta() : false;
// shouldSplitWAL has a default over in pb so this invocation will always work.
this.shouldSplitWal = state.getShouldSplitWal();
int size = state.getRegionsOnCrashedServerCount();
if (size > 0) {
this.regionsOnCrashedServer = new ArrayList<>(size);
for (org.apache.hadoop.hbase.shaded.protobuf.generated.HBaseProtos.RegionInfo ri : state
.getRegionsOnCrashedServerList()) {
this.regionsOnCrashedServer.add(ProtobufUtil.toRegionInfo(ri));
}
}
updateProgress(false);
}
@Override
public ServerName getServerName() {
return this.serverName;
}
@Override
public boolean hasMetaTableRegion() {
return this.carryingMeta;
}
@Override
public ServerOperationType getServerOperationType() {
return ServerOperationType.CRASH_HANDLER;
}
@Override
protected boolean shouldWaitClientAck(MasterProcedureEnv env) {
// The operation is triggered internally on the server
// the client does not know about this procedure.
return false;
}
/**
* Moved out here so can be overridden by the HBCK fix-up SCP to be less strict about what it will
* tolerate as a 'match'.
* @return True if the region location in rsn matches that of this crashed server.
*/
protected boolean isMatchingRegionLocation(RegionStateNode rsn) {
return this.serverName.equals(rsn.getRegionLocation());
}
/**
* Assign the regions on the crashed RS to other Rses.
*
* In this method we will go through all the RegionStateNodes of the give regions to find out
* whether there is already an TRSP for the region, if so we interrupt it and let it retry on
* other server, otherwise we will schedule a TRSP to bring the region online.
*
* We will also check whether the table for a region is enabled, if not, we will skip assigning
* it.
*/
private void assignRegions(MasterProcedureEnv env, List regions) throws IOException {
AssignmentManager am = env.getMasterServices().getAssignmentManager();
boolean retainAssignment = env.getMasterConfiguration().getBoolean(MASTER_SCP_RETAIN_ASSIGNMENT,
DEFAULT_MASTER_SCP_RETAIN_ASSIGNMENT);
for (RegionInfo region : regions) {
RegionStateNode regionNode = am.getRegionStates().getOrCreateRegionStateNode(region);
regionNode.lock();
try {
// This is possible, as when a server is dead, TRSP will fail to schedule a RemoteProcedure
// and then try to assign the region to a new RS. And before it has updated the region
// location to the new RS, we may have already called the am.getRegionsOnServer so we will
// consider the region is still on this crashed server. Then before we arrive here, the
// TRSP could have updated the region location, or even finished itself, so the region is
// no longer on this crashed server any more. We should not try to assign it again. Please
// see HBASE-23594 for more details.
// UPDATE: HBCKServerCrashProcedure overrides isMatchingRegionLocation; this check can get
// in the way of our clearing out 'Unknown Servers'.
if (!isMatchingRegionLocation(regionNode)) {
// See HBASE-24117, though we have already changed the shutdown order, it is still worth
// double checking here to confirm that we do not skip assignment incorrectly.
if (!am.isRunning()) {
throw new DoNotRetryIOException(
"AssignmentManager has been stopped, can not process assignment any more");
}
LOG.info("{} found {} whose regionLocation no longer matches {}, skipping assign...",
this, regionNode, serverName);
continue;
}
if (regionNode.getProcedure() != null) {
LOG.info("{} found RIT {}; {}", this, regionNode.getProcedure(), regionNode);
regionNode.getProcedure().serverCrashed(env, regionNode, getServerName(),
!retainAssignment);
continue;
}
if (
env.getMasterServices().getTableStateManager().isTableState(regionNode.getTable(),
TableState.State.DISABLING)
) {
// We need to change the state here otherwise the TRSP scheduled by DTP will try to
// close the region from a dead server and will never succeed. Please see HBASE-23636
// for more details.
env.getAssignmentManager().regionClosedAbnormally(regionNode);
LOG.info("{} found table disabling for region {}, set it state to ABNORMALLY_CLOSED.",
this, regionNode);
continue;
}
if (
env.getMasterServices().getTableStateManager().isTableState(regionNode.getTable(),
TableState.State.DISABLED)
) {
// This should not happen, table disabled but has regions on server.
LOG.warn("Found table disabled for region {}, procDetails: {}", regionNode, this);
continue;
}
TransitRegionStateProcedure proc =
TransitRegionStateProcedure.assign(env, region, !retainAssignment, null);
regionNode.setProcedure(proc);
addChildProcedure(proc);
} finally {
regionNode.unlock();
}
}
}
@Override
protected ProcedureMetrics getProcedureMetrics(MasterProcedureEnv env) {
return env.getMasterServices().getMasterMetrics().getServerCrashProcMetrics();
}
@Override
protected boolean holdLock(MasterProcedureEnv env) {
return true;
}
public static void updateProgress(MasterProcedureEnv env, long parentId) {
if (parentId == NO_PROC_ID) {
return;
}
Procedure parentProcedure =
env.getMasterServices().getMasterProcedureExecutor().getProcedure(parentId);
if (parentProcedure != null && parentProcedure instanceof ServerCrashProcedure) {
((ServerCrashProcedure) parentProcedure).updateProgress(false);
}
}
}