CEP-15: C* - Early repair failures hang nodetool

Patch by Blake Eggleston; Reviewed by Ariel Weisberg for CASSANDRA-19834
This commit is contained in:
Blake Eggleston 2024-08-13 15:24:10 -07:00 committed by David Capwell
parent 058640be4e
commit 0f51ee407b
6 changed files with 142 additions and 49 deletions

View File

@ -169,6 +169,7 @@ import org.apache.cassandra.schema.TableMetadata;
import org.apache.cassandra.schema.TableMetadataRef;
import org.apache.cassandra.schema.ViewMetadata;
import org.apache.cassandra.service.consensus.migration.ConsensusMigrationState;
import org.apache.cassandra.service.consensus.migration.ConsensusMigrationTarget;
import org.apache.cassandra.service.disk.usage.DiskUsageBroadcaster;
import org.apache.cassandra.service.paxos.Paxos;
import org.apache.cassandra.service.paxos.PaxosCommit;
@ -1686,12 +1687,13 @@ public class StorageService extends NotificationBroadcasterSupport implements IE
}
@Override
public List<Integer> finishConsensusMigration(@Nonnull String keyspace,
@Nullable List<String> maybeTableNames,
@Nullable String maybeRangesStr)
public Integer finishConsensusMigration(@Nonnull String keyspace,
@Nullable List<String> maybeTableNames,
@Nullable String maybeRangesStr,
@Nonnull ConsensusMigrationTarget target)
{
checkArgument(!keyspace.equals(SchemaConstants.METADATA_KEYSPACE_NAME));
return finishMigrationToConsensusProtocol(keyspace, Optional.ofNullable(maybeTableNames), Optional.ofNullable(maybeRangesStr));
return finishMigrationToConsensusProtocol(keyspace, Optional.ofNullable(maybeTableNames), Optional.ofNullable(maybeRangesStr), target);
}
@Override

View File

@ -36,6 +36,7 @@ import javax.management.openmbean.TabularData;
import org.apache.cassandra.db.ColumnFamilyStoreMBean;
import org.apache.cassandra.exceptions.ConfigurationException;
import org.apache.cassandra.service.consensus.migration.ConsensusMigrationTarget;
import org.apache.cassandra.utils.BreaksJMX;
public interface StorageServiceMBean extends NotificationEmitter
@ -1147,9 +1148,10 @@ public interface StorageServiceMBean extends NotificationEmitter
@Nullable List<String> maybeTableNames,
@Nullable String maybeRangesStr);
List<Integer> finishConsensusMigration(@Nonnull String keyspace,
@Nullable List<String> maybeTableNames,
@Nullable String maybeRangesStr);
Integer finishConsensusMigration(@Nonnull String keyspace,
@Nullable List<String> maybeTableNames,
@Nullable String maybeRangesStr,
@Nonnull ConsensusMigrationTarget target);
String listConsensusMigrations(@Nullable Set<String> keyspaceNames, @Nullable Set<String> tableNames, @Nonnull String format);

View File

@ -181,11 +181,13 @@ public abstract class ConsensusTableMigration
ClusterMetadataService.instance().commit(new BeginConsensusMigrationForTableAndRange(targetProtocol, ranges, tableIds));
}
public static List<Integer> finishMigrationToConsensusProtocol(@Nonnull String keyspace,
public static Integer finishMigrationToConsensusProtocol(@Nonnull String keyspace,
@Nonnull Optional<List<String>> maybeTables,
@Nonnull Optional<String> maybeRangesStr)
@Nonnull Optional<String> maybeRangesStr,
ConsensusMigrationTarget target)
{
checkArgument(!maybeTables.isPresent() || !maybeTables.get().isEmpty(), "Must provide at least 1 table if Optional is not empty");
checkNotNull(target);
Optional<List<Range<Token>>> localKeyspaceRanges = Optional.of(ImmutableList.copyOf(StorageService.instance.getLocalReplicas(keyspace).onlyFull().ranges()));
List<Range<Token>> ranges = maybeRangesToRanges(maybeRangesStr, localKeyspaceRanges);
@ -215,17 +217,17 @@ public abstract class ConsensusTableMigration
tableMigrationStates.add(tms);
});
List<TableMigrationState> migratingToAccord = tableMigrationStates.stream().filter(tms -> tms.targetProtocol == ConsensusMigrationTarget.accord).collect(toImmutableList());
List<TableMigrationState> migratingToPaxos = tableMigrationStates.stream().filter(tms -> tms.targetProtocol == ConsensusMigrationTarget.paxos).collect(toImmutableList());;
Integer accordRepairCmd = finishMigrationToAccord(keyspace, migratingToAccord, ranges);
Integer paxosRepairCmd = finishMigrationToPaxos(keyspace, migratingToPaxos, ranges);
List<Integer> result = new ArrayList<>();
if (accordRepairCmd != null)
result.add(accordRepairCmd);
if (paxosRepairCmd != null)
result.add(paxosRepairCmd);
return result;
switch (target)
{
case accord:
List<TableMigrationState> migratingToAccord = tableMigrationStates.stream().filter(tms -> tms.targetProtocol == ConsensusMigrationTarget.accord).collect(toImmutableList());
return finishMigrationToAccord(keyspace, migratingToAccord, ranges);
case paxos:
List<TableMigrationState> migratingToPaxos = tableMigrationStates.stream().filter(tms -> tms.targetProtocol == ConsensusMigrationTarget.paxos).collect(toImmutableList());;
return finishMigrationToPaxos(keyspace, migratingToPaxos, ranges);
default:
throw new IllegalArgumentException("Unsupported target: " + target);
}
}
private interface MigrationFinisher

View File

@ -30,6 +30,7 @@ import java.rmi.server.RMIClientSocketFactory;
import java.rmi.server.RMISocketFactory;
import java.util.AbstractMap;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
@ -114,7 +115,6 @@ import org.apache.cassandra.service.ActiveRepairServiceMBean;
import org.apache.cassandra.service.CacheService;
import org.apache.cassandra.service.CacheServiceMBean;
import org.apache.cassandra.service.snapshot.SnapshotManagerMBean;
import org.apache.cassandra.tcm.CMSOperationsMBean;
import org.apache.cassandra.service.GCInspector;
import org.apache.cassandra.service.GCInspectorMXBean;
import org.apache.cassandra.service.StorageProxy;
@ -126,6 +126,8 @@ import org.apache.cassandra.streaming.StreamManagerMBean;
import org.apache.cassandra.streaming.StreamState;
import org.apache.cassandra.streaming.management.StreamStateCompositeData;
import org.apache.cassandra.tcm.CMSOperations;
import org.apache.cassandra.tcm.CMSOperationsMBean;
import org.apache.cassandra.tools.RepairRunner.RepairCmd;
import org.apache.cassandra.tools.nodetool.GetTimeout;
import org.apache.cassandra.tools.nodetool.formatter.TableBuilder;
import org.apache.cassandra.utils.NativeLibrary;
@ -525,18 +527,28 @@ public class NodeProbe implements AutoCloseable
public void repairAsync(final PrintStream out, final String keyspace, Map<String, String> options) throws IOException
{
blockOnAsyncRepair(out, keyspace, ssProxy.repairAsync(keyspace, options));
startAndBlockOnAsyncRepairs(out, Collections.singleton(new RepairCmd(keyspace)
{
@Override
public Integer start()
{
return ssProxy.repairAsync(keyspace, options);
}
}));
}
public void blockOnAsyncRepair(final PrintStream out, final String keyspace, Integer cmd) throws IOException
public void startAndBlockOnAsyncRepairs(final PrintStream out, Collection<RepairCmd> cmds) throws IOException
{
RepairRunner runner = new RepairRunner(out, ssProxy, keyspace, cmd);
List<RepairRunner> runners = new ArrayList<>(cmds.size());
for (RepairCmd cmd : cmds)
runners.add(new RepairRunner(out, jmxc, ssProxy, cmd));
try
{
if (jmxc != null)
jmxc.addConnectionNotificationListener(runner, null, null);
ssProxy.addNotificationListener(runner, null, null);
runner.run();
runners.forEach(RepairRunner::start);
for (RepairRunner runner : runners)
runner.run();
}
catch (Exception e)
{
@ -546,9 +558,7 @@ public class NodeProbe implements AutoCloseable
{
try
{
ssProxy.removeNotificationListener(runner);
if (jmxc != null)
jmxc.removeConnectionNotificationListener(runner);
runners.forEach(RepairRunner::close);
}
catch (Throwable e)
{

View File

@ -22,6 +22,9 @@ import java.io.PrintStream;
import java.text.SimpleDateFormat;
import java.util.List;
import javax.management.ListenerNotFoundException;
import javax.management.remote.JMXConnector;
import org.apache.cassandra.service.ActiveRepairService.ParentRepairStatus;
import org.apache.cassandra.service.StorageServiceMBean;
import org.apache.cassandra.utils.concurrent.Condition;
@ -42,30 +45,78 @@ import static org.apache.cassandra.utils.progress.ProgressEventType.PROGRESS;
public class RepairRunner extends JMXNotificationProgressListener
{
public static abstract class RepairCmd
{
private final String keyspace;
public RepairCmd(String keyspace)
{
this.keyspace = keyspace;
}
public abstract Integer start();
}
private final SimpleDateFormat format = new SimpleDateFormat("yyyy-MM-dd HH:mm:ss,SSS");
private final PrintStream out;
private final JMXConnector jmxc;
private final StorageServiceMBean ssProxy;
private final String keyspace;
private final Condition condition = newOneTimeCondition();
private final RepairCmd repairCmd;
private Integer cmd;
private volatile Exception error;
public RepairRunner(PrintStream out, StorageServiceMBean ssProxy, String keyspace, Integer cmd)
public RepairRunner(PrintStream out, JMXConnector jmxc, StorageServiceMBean ssProxy, RepairCmd repairCmd)
{
this.out = out;
this.jmxc = jmxc;
this.ssProxy = ssProxy;
this.keyspace = keyspace;
this.cmd = cmd;
this.repairCmd = repairCmd;
}
public void start()
{
if (jmxc != null)
jmxc.addConnectionNotificationListener(this, null, null);
ssProxy.addNotificationListener(this, null, null);
this.cmd = repairCmd.start();
}
public void close()
{
try
{
ssProxy.removeNotificationListener(this);
}
catch (ListenerNotFoundException e)
{
// noop - there may be double removes with error handling
}
if (jmxc != null)
{
try
{
jmxc.removeConnectionNotificationListener(this);
}
catch (ListenerNotFoundException e)
{
// noop - there may be double removes with error handling
}
}
}
public void run() throws Exception
{
if (cmd == null)
return;
if (cmd <= 0)
{
// repairAsync can only return 0 for replication factor 1.
String message = String.format("Replication factor is 1. No repair is needed for keyspace '%s'", keyspace);
String message = String.format("Replication factor is 1. No repair is needed for keyspace '%s'", repairCmd.keyspace);
printMessage(message);
}
else
@ -117,7 +168,7 @@ public class RepairRunner extends JMXNotificationProgressListener
{
error = new IOException(String.format("[%s] JMX connection closed. You should check server log for repair status of keyspace %s"
+ "(Subsequent keyspaces are not going to be repaired).",
format.format(timestamp), keyspace));
format.format(timestamp), repairCmd.keyspace));
condition.signalAll();
}

View File

@ -27,8 +27,10 @@ import java.util.Set;
import io.airlift.airline.Arguments;
import io.airlift.airline.Command;
import io.airlift.airline.Option;
import org.apache.cassandra.service.consensus.migration.ConsensusMigrationTarget;
import org.apache.cassandra.tools.NodeProbe;
import org.apache.cassandra.tools.NodeTool;
import org.apache.cassandra.tools.RepairRunner.RepairCmd;
import static com.google.common.base.Preconditions.checkArgument;
import static java.util.Collections.singleton;
@ -96,26 +98,50 @@ public abstract class ConsensusMigrationAdmin extends NodeTool.NodeToolCmd
@Arguments(usage = "[<keyspace> <tables>...]", description = "The keyspace followed by one or many tables")
private List<String> schemaArgs = new ArrayList<>();
private static class FinishMigrationRepairCommand extends RepairCmd
{
private final NodeProbe probe;
private final String keyspace;
private final List<String> maybeTableNames;
private final String maybeRangesStr;
private final ConsensusMigrationTarget target;
public FinishMigrationRepairCommand(NodeProbe probe, String keyspace, List<String> maybeTableNames, String maybeRangesStr, ConsensusMigrationTarget target)
{
super(keyspace);
this.probe = probe;
this.keyspace = keyspace;
this.maybeTableNames = maybeTableNames;
this.maybeRangesStr = maybeRangesStr;
this.target = target;
}
@Override
public Integer start()
{
return probe.getStorageService().finishConsensusMigration(keyspace, maybeTableNames, maybeRangesStr, target);
}
}
protected void execute(NodeProbe probe)
{
checkArgument((endToken != null) == (startToken != null), "Start and end token must be specified together");
String maybeRangesStr = startToken != null ? startToken + ":" + endToken : null;
List<String> keyspaceNames = parseOptionalKeyspace(schemaArgs, probe, KeyspaceSet.ACCORD_MANAGED);
List<String> maybeTableNames = schemaArgs.size() > 1 ? schemaArgs.subList(1, schemaArgs.size()) : null;
List<RepairCmd> repairCmds = new ArrayList<>(keyspaceNames.size() * 2);
for (String keyspace : keyspaceNames)
{
List<Integer> commands = probe.getStorageService().finishConsensusMigration(keyspace, maybeTableNames, maybeRangesStr);
for (Integer command : commands)
{
try
{
probe.blockOnAsyncRepair(probe.output().out, keyspace, command);
}
catch (IOException e)
{
throw new RuntimeException("Error occurred attempting to finish migration for keyspace " + keyspace + " tables " + maybeTableNames + " and ranges " + maybeRangesStr, e);
}
}
repairCmds.add(new FinishMigrationRepairCommand(probe, keyspace, maybeTableNames, maybeRangesStr, ConsensusMigrationTarget.paxos));
repairCmds.add(new FinishMigrationRepairCommand(probe, keyspace, maybeTableNames, maybeRangesStr, ConsensusMigrationTarget.accord));
}
try
{
probe.startAndBlockOnAsyncRepairs(probe.output().out, repairCmds);
}
catch (IOException e)
{
throw new RuntimeException("Error occurred attempting to finish migration for keyspace(s) " + keyspaceNames + " tables " + maybeTableNames + " and ranges " + maybeRangesStr, e);
}
probe.output().out.printf("Finished consensus migration range (%s) of keyspaces %s and tables %s%n", maybeRangesStr, keyspaceNames, maybeTableNames);
}