private void applyCleanedIndices(final ClusterChangedEvent event) {
// handle closed indices, since they are not allocated on a node once they are closed
// so applyDeletedIndices might not take them into account
for (IndexService indexService : indicesService) {
String index = indexService.index().getName();
IndexMetaData indexMetaData = event.state().metaData().index(index);
if (indexMetaData != null && indexMetaData.state() == IndexMetaData.State.CLOSE) {
for (Integer shardId : indexService.shardIds()) {
logger.debug("[{}][{}] removing shard (index is closed)", index, shardId);
try {
indexService.removeShard(shardId, "removing shard (index is closed)");
} catch (Throwable e) {
logger.warn("[{}] failed to remove shard (index is closed)", e, index);
}
}
}
}
for (IndexService indexService : indicesService) {
String index = indexService.index().getName();
if (indexService.shardIds().isEmpty()) {
if (logger.isDebugEnabled()) {
logger.debug("[{}] cleaning index (no shards allocated)", index);
}
// clean the index
removeIndex(index, "removing index (no shards allocated)");
}
}
}
private void applyDeletedIndices(final ClusterChangedEvent event) {
final ClusterState previousState = event.previousState();
final String localNodeId = event.state().nodes().localNodeId();
assert localNodeId != null;
for (IndexService indexService : indicesService) {
IndexMetaData indexMetaData = event.state().metaData().index(indexService.index().name());
if (indexMetaData != null) {
if (!indexMetaData.isSameUUID(indexService.indexUUID())) {
logger.debug(
"[{}] mismatch on index UUIDs between cluster state and local state, cleaning the index so it will be recreated",
indexMetaData.index());
deleteIndex(
indexMetaData.index(),
"mismatch on index UUIDs between cluster state and local state, cleaning the index so it will be recreated");
}
}
}
for (String index : event.indicesDeleted()) {
if (logger.isDebugEnabled()) {
logger.debug("[{}] cleaning index, no longer part of the metadata", index);
}
final Settings indexSettings;
final IndexService idxService = indicesService.indexService(index);
if (idxService != null) {
indexSettings = idxService.getIndexSettings();
deleteIndex(index, "index no longer part of the metadata");
} else {
final IndexMetaData metaData = previousState.metaData().index(index);
assert metaData != null;
indexSettings = metaData.settings();
indicesService.deleteClosedIndex(
"closed index no longer part of the metadata", metaData, event.state());
}
try {
nodeIndexDeletedAction.nodeIndexDeleted(event.state(), index, indexSettings, localNodeId);
} catch (Throwable e) {
logger.debug("failed to send to master index {} deleted event", e, index);
}
}
}
private void cleanFailedShards(final ClusterChangedEvent event) {
RoutingTable routingTable = event.state().routingTable();
RoutingNodes.RoutingNodeIterator routingNode =
event.state().readOnlyRoutingNodes().routingNodeIter(event.state().nodes().localNodeId());
if (routingNode == null) {
failedShards.clear();
return;
}
DiscoveryNodes nodes = event.state().nodes();
long now = System.currentTimeMillis();
String localNodeId = nodes.localNodeId();
Iterator<Map.Entry<ShardId, FailedShard>> iterator = failedShards.entrySet().iterator();
shards:
while (iterator.hasNext()) {
Map.Entry<ShardId, FailedShard> entry = iterator.next();
FailedShard failedShard = entry.getValue();
IndexRoutingTable indexRoutingTable = routingTable.index(entry.getKey().getIndex());
if (indexRoutingTable != null) {
IndexShardRoutingTable shardRoutingTable = indexRoutingTable.shard(entry.getKey().id());
if (shardRoutingTable != null) {
for (ShardRouting shardRouting : shardRoutingTable.assignedShards()) {
if (localNodeId.equals(shardRouting.currentNodeId())) {
// we have a timeout here just to make sure we don't have dangled failed shards for
// some reason
// its just another safely layer
if (shardRouting.version() == failedShard.version
&& ((now - failedShard.timestamp) < TimeValue.timeValueMinutes(60).millis())) {
// It's the same failed shard - keep it if it hasn't timed out
continue shards;
} else {
// Different version or expired, remove it
break;
}
}
}
}
}
iterator.remove();
}
}
private void applyDeletedShards(final ClusterChangedEvent event) {
RoutingNodes.RoutingNodeIterator routingNode =
event.state().readOnlyRoutingNodes().routingNodeIter(event.state().nodes().localNodeId());
if (routingNode == null) {
return;
}
IntHashSet newShardIds = new IntHashSet();
for (IndexService indexService : indicesService) {
String index = indexService.index().name();
IndexMetaData indexMetaData = event.state().metaData().index(index);
if (indexMetaData == null) {
continue;
}
// now, go over and delete shards that needs to get deleted
newShardIds.clear();
for (ShardRouting shard : routingNode) {
if (shard.index().equals(index)) {
newShardIds.add(shard.id());
}
}
for (Integer existingShardId : indexService.shardIds()) {
if (!newShardIds.contains(existingShardId)) {
if (indexMetaData.state() == IndexMetaData.State.CLOSE) {
if (logger.isDebugEnabled()) {
logger.debug("[{}][{}] removing shard (index is closed)", index, existingShardId);
}
indexService.removeShard(existingShardId, "removing shard (index is closed)");
} else {
// we can just remove the shard, without cleaning it locally, since we will clean it
// when all shards are allocated in the IndicesStore
if (logger.isDebugEnabled()) {
logger.debug("[{}][{}] removing shard (not allocated)", index, existingShardId);
}
indexService.removeShard(existingShardId, "removing shard (not allocated)");
}
}
}
}
}
@Override
public void clusterChanged(ClusterChangedEvent event) {
if (!VALID_CLUSTER_STATES.contains(event.state().status())) {
// Only process fully applied cluster states
return;
}
if (eventBus != null) {
final List<String> indicesDeleted = event.indicesDeleted();
if (!indicesDeleted.isEmpty()) {
eventBus.post(IndicesDeletedEvent.create(indicesDeleted));
}
}
}
private void applyNewIndices(final ClusterChangedEvent event) {
// we only create indices for shards that are allocated
RoutingNodes.RoutingNodeIterator routingNode =
event.state().readOnlyRoutingNodes().routingNodeIter(event.state().nodes().localNodeId());
if (routingNode == null) {
return;
}
for (ShardRouting shard : routingNode) {
if (!indicesService.hasIndex(shard.index())) {
final IndexMetaData indexMetaData = event.state().metaData().index(shard.index());
if (logger.isDebugEnabled()) {
logger.debug("[{}] creating index", indexMetaData.index());
}
try {
indicesService.createIndex(
indexMetaData.index(),
indexMetaData.settings(),
event.state().nodes().localNode().id());
} catch (Throwable e) {
sendFailShard(shard, indexMetaData.getIndexUUID(), "failed to create index", e);
}
}
}
}
@Override
public void clusterChanged(ClusterChangedEvent event) {
if (event.state().blocks().disableStatePersistence()) {
// reset the current metadata, we need to start fresh...
this.currentMetaData = null;
return;
}
MetaData newMetaData = event.state().metaData();
// delete indices that were there before, but are deleted now
// we need to do it so they won't be detected as dangling
if (currentMetaData != null) {
// only delete indices when we already received a state (currentMetaData != null)
for (IndexMetaData current : currentMetaData) {
if (!newMetaData.hasIndex(current.index())) {
logger.debug(
"[{}] deleting index that is no longer part of the metadata (indices: [{}])",
current.index(),
newMetaData.indices().keys());
if (nodeEnv.hasNodeFile()) {
FileSystemUtils.deleteRecursively(nodeEnv.indexLocations(new Index(current.index())));
}
try {
nodeIndexDeletedAction.nodeIndexStoreDeleted(
event.state(), current.index(), event.state().nodes().localNodeId());
} catch (Exception e) {
logger.debug(
"[{}] failed to notify master on local index store deletion", e, current.index());
}
}
}
}
currentMetaData = newMetaData;
}
private void applyAliases(ClusterChangedEvent event) {
// check if aliases changed
if (aliasesChanged(event)) {
// go over and update aliases
for (IndexMetaData indexMetaData : event.state().metaData()) {
String index = indexMetaData.index();
IndexService indexService = indicesService.indexService(index);
if (indexService == null) {
// we only create / update here
continue;
}
IndexAliasesService indexAliasesService = indexService.aliasesService();
indexAliasesService.setAliases(indexMetaData.getAliases());
}
}
}
@Override
public void clusterChanged(final ClusterChangedEvent event) {
if (!indicesService.changesAllowed()) {
return;
}
if (!lifecycle.started()) {
return;
}
synchronized (mutex) {
// we need to clean the shards and indices we have on this node, since we
// are going to recover them again once state persistence is disabled (no master / not
// recovered)
// TODO: this feels a bit hacky here, a block disables state persistence, and then we clean
// the allocated shards, maybe another flag in blocks?
if (event.state().blocks().disableStatePersistence()) {
for (IndexService indexService : indicesService) {
String index = indexService.index().getName();
for (Integer shardId : indexService.shardIds()) {
logger.debug("[{}][{}] removing shard (disabled block persistence)", index, shardId);
try {
indexService.removeShard(shardId, "removing shard (disabled block persistence)");
} catch (Throwable e) {
logger.warn("[{}] failed to remove shard (disabled block persistence)", e, index);
}
}
removeIndex(index, "cleaning index (disabled block persistence)");
}
return;
}
cleanFailedShards(event);
applyDeletedIndices(event);
applyNewIndices(event);
applyMappings(event);
applyAliases(event);
applyNewOrUpdatedShards(event);
applyDeletedShards(event);
applyCleanedIndices(event);
applySettings(event);
}
}
@Override
public void clusterChanged(ClusterChangedEvent event) {
if (event.source().equals(CLUSTER_UPDATE_TASK_SOURCE)) {
// that's us, ignore this event
return;
}
if (event.state().nodes().localNodeMaster()) {
// we are master, schedule the routing table updater
if (scheduledRoutingTableFuture == null) {
// a new master (us), make sure we reroute shards
routingTableDirty = true;
scheduledRoutingTableFuture =
threadPool.scheduleWithFixedDelay(new RoutingTableUpdater(), schedule);
}
if (event.nodesRemoved()) {
// if nodes were removed, we don't want to wait for the scheduled task
// since we want to get primary election as fast as possible
routingTableDirty = true;
reroute();
// Commented out since we make sure to reroute whenever shards changes state or metadata
// changes state
// } else if (event.routingTableChanged()) {
// routingTableDirty = true;
// reroute();
} else {
if (event.nodesAdded()) {
for (DiscoveryNode node : event.nodesDelta().addedNodes()) {
if (node.dataNode()) {
routingTableDirty = true;
break;
}
}
}
}
} else {
FutureUtils.cancel(scheduledRoutingTableFuture);
scheduledRoutingTableFuture = null;
}
}
private void applySettings(ClusterChangedEvent event) {
if (!event.metaDataChanged()) {
return;
}
for (IndexMetaData indexMetaData : event.state().metaData()) {
if (!indicesService.hasIndex(indexMetaData.index())) {
// we only create / update here
continue;
}
// if the index meta data didn't change, no need check for refreshed settings
if (!event.indexMetaDataChanged(indexMetaData)) {
continue;
}
String index = indexMetaData.index();
IndexService indexService = indicesService.indexService(index);
if (indexService == null) {
// already deleted on us, ignore it
continue;
}
IndexSettingsService indexSettingsService =
indexService.injector().getInstance(IndexSettingsService.class);
indexSettingsService.refreshSettings(indexMetaData.settings());
}
}
private void applyNewOrUpdatedShards(final ClusterChangedEvent event) {
if (!indicesService.changesAllowed()) {
return;
}
RoutingTable routingTable = event.state().routingTable();
RoutingNodes.RoutingNodeIterator routingNode =
event.state().readOnlyRoutingNodes().routingNodeIter(event.state().nodes().localNodeId());
if (routingNode == null) {
failedShards.clear();
return;
}
DiscoveryNodes nodes = event.state().nodes();
for (final ShardRouting shardRouting : routingNode) {
final IndexService indexService = indicesService.indexService(shardRouting.index());
if (indexService == null) {
// got deleted on us, ignore
continue;
}
final IndexMetaData indexMetaData = event.state().metaData().index(shardRouting.index());
if (indexMetaData == null) {
// the index got deleted on the metadata, we will clean it later in the apply deleted method
// call
continue;
}
final int shardId = shardRouting.id();
if (!indexService.hasShard(shardId) && shardRouting.started()) {
if (failedShards.containsKey(shardRouting.shardId())) {
if (nodes.masterNode() != null) {
shardStateAction.resendShardFailed(
shardRouting,
indexMetaData.getIndexUUID(),
"master "
+ nodes.masterNode()
+ " marked shard as started, but shard has previous failed. resending shard failure.",
nodes.masterNode());
}
} else {
// the master thinks we are started, but we don't have this shard at all, mark it as
// failed
sendFailShard(
shardRouting,
indexMetaData.getIndexUUID(),
"master ["
+ nodes.masterNode()
+ "] marked shard as started, but shard has not been created, mark shard as failed",
null);
}
continue;
}
IndexShard indexShard = indexService.shard(shardId);
if (indexShard != null) {
ShardRouting currentRoutingEntry = indexShard.routingEntry();
// if the current and global routing are initializing, but are still not the same, its a
// different "shard" being allocated
// for example: a shard that recovers from one node and now needs to recover to another
// node,
// or a replica allocated and then allocating a primary because the primary
// failed on another node
boolean shardHasBeenRemoved = false;
if (currentRoutingEntry.initializing()
&& shardRouting.initializing()
&& !currentRoutingEntry.equals(shardRouting)) {
logger.debug(
"[{}][{}] removing shard (different instance of it allocated on this node, current [{}], global [{}])",
shardRouting.index(),
shardRouting.id(),
currentRoutingEntry,
shardRouting);
// closing the shard will also cancel any ongoing recovery.
indexService.removeShard(
shardRouting.id(),
"removing shard (different instance of it allocated on this node)");
shardHasBeenRemoved = true;
} else if (isPeerRecovery(shardRouting)) {
final DiscoveryNode sourceNode =
findSourceNodeForPeerRecovery(routingTable, nodes, shardRouting);
// check if there is an existing recovery going, and if so, and the source node is not the
// same, cancel the recovery to restart it
final Predicate<RecoveryStatus> shouldCancel =
new Predicate<RecoveryStatus>() {
@Override
public boolean apply(@Nullable RecoveryStatus status) {
return status.sourceNode().equals(sourceNode) == false;
}
};
if (recoveryTarget.cancelRecoveriesForShard(
indexShard.shardId(), "recovery source node changed", shouldCancel)) {
logger.debug(
"[{}][{}] removing shard (recovery source changed), current [{}], global [{}])",
shardRouting.index(),
shardRouting.id(),
currentRoutingEntry,
shardRouting);
// closing the shard will also cancel any ongoing recovery.
indexService.removeShard(
shardRouting.id(), "removing shard (recovery source node changed)");
shardHasBeenRemoved = true;
}
}
if (shardHasBeenRemoved == false
&& (shardRouting.equals(indexShard.routingEntry()) == false
|| shardRouting.version() > indexShard.routingEntry().version())) {
if (shardRouting.primary()
&& indexShard.routingEntry().primary() == false
&& shardRouting.initializing()
&& indexShard.allowsPrimaryPromotion() == false) {
logger.debug("{} reinitialize shard on primary promotion", indexShard.shardId());
indexService.removeShard(shardId, "promoted to primary");
} else {
// if we happen to remove the shardRouting by id above we don't need to jump in here!
indexShard.updateRoutingEntry(
shardRouting, event.state().blocks().disableStatePersistence() == false);
}
}
}
if (shardRouting.initializing()) {
applyInitializingShard(event.state(), indexMetaData, shardRouting);
}
}
}
private boolean aliasesChanged(ClusterChangedEvent event) {
return !event.state().metaData().aliases().equals(event.previousState().metaData().aliases())
|| !event.state().routingTable().equals(event.previousState().routingTable());
}
private void applyMappings(ClusterChangedEvent event) {
// go over and update mappings
for (IndexMetaData indexMetaData : event.state().metaData()) {
if (!indicesService.hasIndex(indexMetaData.index())) {
// we only create / update here
continue;
}
List<String> typesToRefresh = Lists.newArrayList();
String index = indexMetaData.index();
IndexService indexService = indicesService.indexService(index);
if (indexService == null) {
// got deleted on us, ignore (closing the node)
return;
}
try {
MapperService mapperService = indexService.mapperService();
// first, go over and update the _default_ mapping (if exists)
if (indexMetaData.mappings().containsKey(MapperService.DEFAULT_MAPPING)) {
boolean requireRefresh =
processMapping(
index,
mapperService,
MapperService.DEFAULT_MAPPING,
indexMetaData.mapping(MapperService.DEFAULT_MAPPING).source());
if (requireRefresh) {
typesToRefresh.add(MapperService.DEFAULT_MAPPING);
}
}
// go over and add the relevant mappings (or update them)
for (ObjectCursor<MappingMetaData> cursor : indexMetaData.mappings().values()) {
MappingMetaData mappingMd = cursor.value;
String mappingType = mappingMd.type();
CompressedXContent mappingSource = mappingMd.source();
if (mappingType.equals(MapperService.DEFAULT_MAPPING)) { // we processed _default_ first
continue;
}
boolean requireRefresh = processMapping(index, mapperService, mappingType, mappingSource);
if (requireRefresh) {
typesToRefresh.add(mappingType);
}
}
if (!typesToRefresh.isEmpty() && sendRefreshMapping) {
nodeMappingRefreshAction.nodeMappingRefresh(
event.state(),
new NodeMappingRefreshAction.NodeMappingRefreshRequest(
index,
indexMetaData.indexUUID(),
typesToRefresh.toArray(new String[typesToRefresh.size()]),
event.state().nodes().localNodeId()));
}
} catch (Throwable t) {
// if we failed the mappings anywhere, we need to fail the shards for this index, note, we
// safeguard
// by creating the processing the mappings on the master, or on the node the mapping was
// introduced on,
// so this failure typically means wrong node level configuration or something similar
for (IndexShard indexShard : indexService) {
ShardRouting shardRouting = indexShard.routingEntry();
failAndRemoveShard(shardRouting, indexService, true, "failed to update mappings", t);
}
}
}
}
private ClusterState applyUpdate(ClusterState currentState, ClusterChangedEvent task) {
boolean clusterStateChanged = false;
ClusterState tribeState = task.state();
DiscoveryNodes.Builder nodes = DiscoveryNodes.builder(currentState.nodes());
// -- merge nodes
// go over existing nodes, and see if they need to be removed
for (DiscoveryNode discoNode : currentState.nodes()) {
String markedTribeName = discoNode.attributes().get(TRIBE_NAME);
if (markedTribeName != null && markedTribeName.equals(tribeName)) {
if (tribeState.nodes().get(discoNode.id()) == null) {
clusterStateChanged = true;
logger.info("[{}] removing node [{}]", tribeName, discoNode);
nodes.remove(discoNode.id());
}
}
}
// go over tribe nodes, and see if they need to be added
for (DiscoveryNode tribe : tribeState.nodes()) {
if (currentState.nodes().get(tribe.id()) == null) {
// a new node, add it, but also add the tribe name to the attributes
Map<String, String> tribeAttr = new HashMap<>();
for (ObjectObjectCursor<String, String> attr : tribe.attributes()) {
tribeAttr.put(attr.key, attr.value);
}
tribeAttr.put(TRIBE_NAME, tribeName);
DiscoveryNode discoNode =
new DiscoveryNode(
tribe.name(),
tribe.id(),
tribe.getHostName(),
tribe.getHostAddress(),
tribe.address(),
unmodifiableMap(tribeAttr),
tribe.version());
clusterStateChanged = true;
logger.info("[{}] adding node [{}]", tribeName, discoNode);
nodes.put(discoNode);
}
}
// -- merge metadata
ClusterBlocks.Builder blocks = ClusterBlocks.builder().blocks(currentState.blocks());
MetaData.Builder metaData = MetaData.builder(currentState.metaData());
RoutingTable.Builder routingTable = RoutingTable.builder(currentState.routingTable());
// go over existing indices, and see if they need to be removed
for (IndexMetaData index : currentState.metaData()) {
String markedTribeName = index.getSettings().get(TRIBE_NAME);
if (markedTribeName != null && markedTribeName.equals(tribeName)) {
IndexMetaData tribeIndex = tribeState.metaData().index(index.getIndex());
clusterStateChanged = true;
if (tribeIndex == null || tribeIndex.getState() == IndexMetaData.State.CLOSE) {
logger.info("[{}] removing index [{}]", tribeName, index.getIndex());
removeIndex(blocks, metaData, routingTable, index);
} else {
// always make sure to update the metadata and routing table, in case
// there are changes in them (new mapping, shards moving from initializing to started)
routingTable.add(tribeState.routingTable().index(index.getIndex()));
Settings tribeSettings =
Settings.builder().put(tribeIndex.getSettings()).put(TRIBE_NAME, tribeName).build();
metaData.put(IndexMetaData.builder(tribeIndex).settings(tribeSettings));
}
}
}
// go over tribe one, and see if they need to be added
for (IndexMetaData tribeIndex : tribeState.metaData()) {
// if there is no routing table yet, do nothing with it...
IndexRoutingTable table = tribeState.routingTable().index(tribeIndex.getIndex());
if (table == null) {
continue;
}
final IndexMetaData indexMetaData = currentState.metaData().index(tribeIndex.getIndex());
if (indexMetaData == null) {
if (!droppedIndices.contains(tribeIndex.getIndex())) {
// a new index, add it, and add the tribe name as a setting
clusterStateChanged = true;
logger.info("[{}] adding index [{}]", tribeName, tribeIndex.getIndex());
addNewIndex(tribeState, blocks, metaData, routingTable, tribeIndex);
}
} else {
String existingFromTribe = indexMetaData.getSettings().get(TRIBE_NAME);
if (!tribeName.equals(existingFromTribe)) {
// we have a potential conflict on index names, decide what to do...
if (ON_CONFLICT_ANY.equals(onConflict)) {
// we chose any tribe, carry on
} else if (ON_CONFLICT_DROP.equals(onConflict)) {
// drop the indices, there is a conflict
clusterStateChanged = true;
logger.info(
"[{}] dropping index [{}] due to conflict with [{}]",
tribeName,
tribeIndex.getIndex(),
existingFromTribe);
removeIndex(blocks, metaData, routingTable, tribeIndex);
droppedIndices.add(tribeIndex.getIndex());
} else if (onConflict.startsWith(ON_CONFLICT_PREFER)) {
// on conflict, prefer a tribe...
String preferredTribeName = onConflict.substring(ON_CONFLICT_PREFER.length());
if (tribeName.equals(preferredTribeName)) {
// the new one is hte preferred one, replace...
clusterStateChanged = true;
logger.info(
"[{}] adding index [{}], preferred over [{}]",
tribeName,
tribeIndex.getIndex(),
existingFromTribe);
removeIndex(blocks, metaData, routingTable, tribeIndex);
addNewIndex(tribeState, blocks, metaData, routingTable, tribeIndex);
} // else: either the existing one is the preferred one, or we haven't seen one, carry
// on
}
}
}
}
if (!clusterStateChanged) {
return currentState;
} else {
return ClusterState.builder(currentState)
.incrementVersion()
.blocks(blocks)
.nodes(nodes)
.metaData(metaData)
.routingTable(routingTable.build())
.build();
}
}
<T> void runTasksForExecutor(ClusterStateTaskExecutor<T> executor) {
final ArrayList<UpdateTask<T>> toExecute = new ArrayList<>();
final Map<String, ArrayList<T>> processTasksBySource = new HashMap<>();
synchronized (updateTasksPerExecutor) {
List<UpdateTask> pending = updateTasksPerExecutor.remove(executor);
if (pending != null) {
for (UpdateTask<T> task : pending) {
if (task.processed.getAndSet(true) == false) {
logger.trace("will process {}", task.toString(executor));
toExecute.add(task);
processTasksBySource
.computeIfAbsent(task.source, s -> new ArrayList<>())
.add(task.task);
} else {
logger.trace("skipping {}, already processed", task.toString(executor));
}
}
}
}
if (toExecute.isEmpty()) {
return;
}
final String tasksSummary =
processTasksBySource
.entrySet()
.stream()
.map(
entry -> {
String tasks = executor.describeTasks(entry.getValue());
return tasks.isEmpty() ? entry.getKey() : entry.getKey() + "[" + tasks + "]";
})
.reduce((s1, s2) -> s1 + ", " + s2)
.orElse("");
if (!lifecycle.started()) {
logger.debug("processing [{}]: ignoring, cluster_service not started", tasksSummary);
return;
}
logger.debug("processing [{}]: execute", tasksSummary);
ClusterState previousClusterState = clusterState;
if (!previousClusterState.nodes().isLocalNodeElectedMaster() && executor.runOnlyOnMaster()) {
logger.debug("failing [{}]: local node is no longer master", tasksSummary);
toExecute.stream().forEach(task -> task.listener.onNoLongerMaster(task.source));
return;
}
ClusterStateTaskExecutor.BatchResult<T> batchResult;
long startTimeNS = currentTimeInNanos();
try {
List<T> inputs =
toExecute.stream().map(tUpdateTask -> tUpdateTask.task).collect(Collectors.toList());
batchResult = executor.execute(previousClusterState, inputs);
} catch (Exception e) {
TimeValue executionTime =
TimeValue.timeValueMillis(
Math.max(0, TimeValue.nsecToMSec(currentTimeInNanos() - startTimeNS)));
if (logger.isTraceEnabled()) {
logger.trace(
(Supplier<?>)
() ->
new ParameterizedMessage(
"failed to execute cluster state update in [{}], state:\nversion [{}], source [{}]\n{}{}{}",
executionTime,
previousClusterState.version(),
tasksSummary,
previousClusterState.nodes().prettyPrint(),
previousClusterState.routingTable().prettyPrint(),
previousClusterState.getRoutingNodes().prettyPrint()),
e);
}
warnAboutSlowTaskIfNeeded(executionTime, tasksSummary);
batchResult =
ClusterStateTaskExecutor.BatchResult.<T>builder()
.failures(toExecute.stream().map(updateTask -> updateTask.task)::iterator, e)
.build(previousClusterState);
}
assert batchResult.executionResults != null;
assert batchResult.executionResults.size() == toExecute.size()
: String.format(
Locale.ROOT,
"expected [%d] task result%s but was [%d]",
toExecute.size(),
toExecute.size() == 1 ? "" : "s",
batchResult.executionResults.size());
boolean assertsEnabled = false;
assert (assertsEnabled = true);
if (assertsEnabled) {
for (UpdateTask<T> updateTask : toExecute) {
assert batchResult.executionResults.containsKey(updateTask.task)
: "missing task result for " + updateTask.toString(executor);
}
}
ClusterState newClusterState = batchResult.resultingState;
final ArrayList<UpdateTask<T>> proccessedListeners = new ArrayList<>();
// fail all tasks that have failed and extract those that are waiting for results
for (UpdateTask<T> updateTask : toExecute) {
assert batchResult.executionResults.containsKey(updateTask.task)
: "missing " + updateTask.toString(executor);
final ClusterStateTaskExecutor.TaskResult executionResult =
batchResult.executionResults.get(updateTask.task);
executionResult.handle(
() -> proccessedListeners.add(updateTask),
ex -> {
logger.debug(
(Supplier<?>)
() ->
new ParameterizedMessage(
"cluster state update task {} failed", updateTask.toString(executor)),
ex);
updateTask.listener.onFailure(updateTask.source, ex);
});
}
if (previousClusterState == newClusterState) {
for (UpdateTask<T> task : proccessedListeners) {
if (task.listener instanceof AckedClusterStateTaskListener) {
// no need to wait for ack if nothing changed, the update can be counted as acknowledged
((AckedClusterStateTaskListener) task.listener).onAllNodesAcked(null);
}
task.listener.clusterStateProcessed(task.source, previousClusterState, newClusterState);
}
TimeValue executionTime =
TimeValue.timeValueMillis(
Math.max(0, TimeValue.nsecToMSec(currentTimeInNanos() - startTimeNS)));
logger.debug(
"processing [{}]: took [{}] no change in cluster_state", tasksSummary, executionTime);
warnAboutSlowTaskIfNeeded(executionTime, tasksSummary);
return;
}
try {
ArrayList<Discovery.AckListener> ackListeners = new ArrayList<>();
if (newClusterState.nodes().isLocalNodeElectedMaster()) {
// only the master controls the version numbers
Builder builder = ClusterState.builder(newClusterState).incrementVersion();
if (previousClusterState.routingTable() != newClusterState.routingTable()) {
builder.routingTable(
RoutingTable.builder(newClusterState.routingTable())
.version(newClusterState.routingTable().version() + 1)
.build());
}
if (previousClusterState.metaData() != newClusterState.metaData()) {
builder.metaData(
MetaData.builder(newClusterState.metaData())
.version(newClusterState.metaData().version() + 1));
}
newClusterState = builder.build();
for (UpdateTask<T> task : proccessedListeners) {
if (task.listener instanceof AckedClusterStateTaskListener) {
final AckedClusterStateTaskListener ackedListener =
(AckedClusterStateTaskListener) task.listener;
if (ackedListener.ackTimeout() == null || ackedListener.ackTimeout().millis() == 0) {
ackedListener.onAckTimeout();
} else {
try {
ackListeners.add(
new AckCountDownListener(
ackedListener,
newClusterState.version(),
newClusterState.nodes(),
threadPool));
} catch (EsRejectedExecutionException ex) {
if (logger.isDebugEnabled()) {
logger.debug(
"Couldn't schedule timeout thread - node might be shutting down", ex);
}
// timeout straightaway, otherwise we could wait forever as the timeout thread has
// not started
ackedListener.onAckTimeout();
}
}
}
}
}
final Discovery.AckListener ackListener = new DelegetingAckListener(ackListeners);
newClusterState.status(ClusterState.ClusterStateStatus.BEING_APPLIED);
if (logger.isTraceEnabled()) {
logger.trace(
"cluster state updated, source [{}]\n{}", tasksSummary, newClusterState.prettyPrint());
} else if (logger.isDebugEnabled()) {
logger.debug(
"cluster state updated, version [{}], source [{}]",
newClusterState.version(),
tasksSummary);
}
ClusterChangedEvent clusterChangedEvent =
new ClusterChangedEvent(tasksSummary, newClusterState, previousClusterState);
// new cluster state, notify all listeners
final DiscoveryNodes.Delta nodesDelta = clusterChangedEvent.nodesDelta();
if (nodesDelta.hasChanges() && logger.isInfoEnabled()) {
String summary = nodesDelta.shortSummary();
if (summary.length() > 0) {
logger.info("{}, reason: {}", summary, tasksSummary);
}
}
nodeConnectionsService.connectToAddedNodes(clusterChangedEvent);
// if we are the master, publish the new state to all nodes
// we publish here before we send a notification to all the listeners, since if it fails
// we don't want to notify
if (newClusterState.nodes().isLocalNodeElectedMaster()) {
logger.debug("publishing cluster state version [{}]", newClusterState.version());
try {
clusterStatePublisher.accept(clusterChangedEvent, ackListener);
} catch (Discovery.FailedToCommitClusterStateException t) {
final long version = newClusterState.version();
logger.warn(
(Supplier<?>)
() ->
new ParameterizedMessage(
"failing [{}]: failed to commit cluster state version [{}]",
tasksSummary,
version),
t);
proccessedListeners.forEach(task -> task.listener.onFailure(task.source, t));
return;
}
}
// update the current cluster state
clusterState = newClusterState;
logger.debug("set local cluster state to version {}", newClusterState.version());
try {
// nothing to do until we actually recover from the gateway or any other block indicates we
// need to disable persistency
if (clusterChangedEvent.state().blocks().disableStatePersistence() == false
&& clusterChangedEvent.metaDataChanged()) {
final Settings incomingSettings = clusterChangedEvent.state().metaData().settings();
clusterSettings.applySettings(incomingSettings);
}
} catch (Exception ex) {
logger.warn("failed to apply cluster settings", ex);
}
for (ClusterStateListener listener : preAppliedListeners) {
try {
listener.clusterChanged(clusterChangedEvent);
} catch (Exception ex) {
logger.warn("failed to notify ClusterStateListener", ex);
}
}
nodeConnectionsService.disconnectFromRemovedNodes(clusterChangedEvent);
newClusterState.status(ClusterState.ClusterStateStatus.APPLIED);
for (ClusterStateListener listener : postAppliedListeners) {
try {
listener.clusterChanged(clusterChangedEvent);
} catch (Exception ex) {
logger.warn("failed to notify ClusterStateListener", ex);
}
}
// manual ack only from the master at the end of the publish
if (newClusterState.nodes().isLocalNodeElectedMaster()) {
try {
ackListener.onNodeAck(newClusterState.nodes().getLocalNode(), null);
} catch (Exception e) {
final DiscoveryNode localNode = newClusterState.nodes().getLocalNode();
logger.debug(
(Supplier<?>)
() ->
new ParameterizedMessage(
"error while processing ack for master node [{}]", localNode),
e);
}
}
for (UpdateTask<T> task : proccessedListeners) {
task.listener.clusterStateProcessed(task.source, previousClusterState, newClusterState);
}
try {
executor.clusterStatePublished(clusterChangedEvent);
} catch (Exception e) {
logger.error(
(Supplier<?>)
() ->
new ParameterizedMessage(
"exception thrown while notifying executor of new cluster state publication [{}]",
tasksSummary),
e);
}
TimeValue executionTime =
TimeValue.timeValueMillis(
Math.max(0, TimeValue.nsecToMSec(currentTimeInNanos() - startTimeNS)));
logger.debug(
"processing [{}]: took [{}] done applying updated cluster_state (version: {}, uuid: {})",
tasksSummary,
executionTime,
newClusterState.version(),
newClusterState.stateUUID());
warnAboutSlowTaskIfNeeded(executionTime, tasksSummary);
} catch (Exception e) {
TimeValue executionTime =
TimeValue.timeValueMillis(
Math.max(0, TimeValue.nsecToMSec(currentTimeInNanos() - startTimeNS)));
final long version = newClusterState.version();
final String stateUUID = newClusterState.stateUUID();
final String prettyPrint = newClusterState.prettyPrint();
logger.warn(
(Supplier<?>)
() ->
new ParameterizedMessage(
"failed to apply updated cluster state in [{}]:\nversion [{}], uuid [{}], source [{}]\n{}",
executionTime,
version,
stateUUID,
tasksSummary,
prettyPrint),
e);
// TODO: do we want to call updateTask.onFailure here?
}
}
@Override
public void clusterChanged(final ClusterChangedEvent event) {
if (lifecycle.stoppedOrClosed()) {
return;
}
final ClusterState state = event.state();
if (state.nodes().isLocalNodeElectedMaster() == false) {
// not our job to recover
return;
}
if (state.blocks().hasGlobalBlock(STATE_NOT_RECOVERED_BLOCK) == false) {
// already recovered
return;
}
DiscoveryNodes nodes = state.nodes();
if (state.nodes().getMasterNodeId() == null) {
logger.debug("not recovering from gateway, no master elected yet");
} else if (recoverAfterNodes != -1
&& (nodes.getMasterAndDataNodes().size()) < recoverAfterNodes) {
logger.debug(
"not recovering from gateway, nodes_size (data+master) [{}] < recover_after_nodes [{}]",
nodes.getMasterAndDataNodes().size(),
recoverAfterNodes);
} else if (recoverAfterDataNodes != -1 && nodes.getDataNodes().size() < recoverAfterDataNodes) {
logger.debug(
"not recovering from gateway, nodes_size (data) [{}] < recover_after_data_nodes [{}]",
nodes.getDataNodes().size(),
recoverAfterDataNodes);
} else if (recoverAfterMasterNodes != -1
&& nodes.getMasterNodes().size() < recoverAfterMasterNodes) {
logger.debug(
"not recovering from gateway, nodes_size (master) [{}] < recover_after_master_nodes [{}]",
nodes.getMasterNodes().size(),
recoverAfterMasterNodes);
} else {
boolean enforceRecoverAfterTime;
String reason;
if (expectedNodes == -1 && expectedMasterNodes == -1 && expectedDataNodes == -1) {
// no expected is set, honor the setting if they are there
enforceRecoverAfterTime = true;
reason = "recover_after_time was set to [" + recoverAfterTime + "]";
} else {
// one of the expected is set, see if all of them meet the need, and ignore the timeout in
// this case
enforceRecoverAfterTime = false;
reason = "";
if (expectedNodes != -1
&& (nodes.getMasterAndDataNodes().size()
< expectedNodes)) { // does not meet the expected...
enforceRecoverAfterTime = true;
reason =
"expecting ["
+ expectedNodes
+ "] nodes, but only have ["
+ nodes.getMasterAndDataNodes().size()
+ "]";
} else if (expectedDataNodes != -1
&& (nodes.getDataNodes().size() < expectedDataNodes)) { // does not meet the expected...
enforceRecoverAfterTime = true;
reason =
"expecting ["
+ expectedDataNodes
+ "] data nodes, but only have ["
+ nodes.getDataNodes().size()
+ "]";
} else if (expectedMasterNodes != -1
&& (nodes.getMasterNodes().size()
< expectedMasterNodes)) { // does not meet the expected...
enforceRecoverAfterTime = true;
reason =
"expecting ["
+ expectedMasterNodes
+ "] master nodes, but only have ["
+ nodes.getMasterNodes().size()
+ "]";
}
}
performStateRecovery(enforceRecoverAfterTime, reason);
}
}
@Override
public void clusterChanged(ClusterChangedEvent event) {
if (event.state().blocks().disableStatePersistence()) {
// reset the current metadata, we need to start fresh...
this.currentMetaData = null;
return;
}
MetaData newMetaData = event.state().metaData();
// we don't check if metaData changed, since we might be called several times and we need to
// check dangling...
boolean success = true;
// only applied to master node, writing the global and index level states
if (event.state().nodes().localNode().masterNode()) {
// check if the global state changed?
if (currentMetaData == null || !MetaData.isGlobalStateEquals(currentMetaData, newMetaData)) {
try {
writeGlobalState("changed", newMetaData, currentMetaData);
} catch (Exception e) {
success = false;
}
}
// check and write changes in indices
for (IndexMetaData indexMetaData : newMetaData) {
String writeReason = null;
IndexMetaData currentIndexMetaData;
if (currentMetaData == null) {
// a new event..., check from the state stored
currentIndexMetaData = loadIndex(indexMetaData.index());
} else {
currentIndexMetaData = currentMetaData.index(indexMetaData.index());
}
if (currentIndexMetaData == null) {
writeReason = "freshly created";
} else if (currentIndexMetaData.version() != indexMetaData.version()) {
writeReason =
"version changed from ["
+ currentIndexMetaData.version()
+ "] to ["
+ indexMetaData.version()
+ "]";
}
// we update the writeReason only if we really need to write it
if (writeReason == null) {
continue;
}
try {
writeIndex(writeReason, indexMetaData, currentIndexMetaData);
} catch (Exception e) {
success = false;
}
}
}
// delete indices that were there before, but are deleted now
// we need to do it so they won't be detected as dangling
if (nodeEnv.hasNodeFile()) {
if (currentMetaData != null) {
// only delete indices when we already received a state (currentMetaData != null)
// and we had a go at processing dangling indices at least once
// this will also delete the _state of the index itself
for (IndexMetaData current : currentMetaData) {
if (danglingIndices.containsKey(current.index())) {
continue;
}
if (!newMetaData.hasIndex(current.index())) {
logger.debug(
"[{}] deleting index that is no longer part of the metadata (indices: [{}])",
current.index(),
newMetaData.indices().keySet());
FileSystemUtils.deleteRecursively(nodeEnv.indexLocations(new Index(current.index())));
}
}
}
}
// handle dangling indices, we handle those for all nodes that have a node file (data or master)
if (nodeEnv.hasNodeFile()) {
if (danglingTimeout.millis() >= 0) {
synchronized (danglingMutex) {
for (String danglingIndex : danglingIndices.keySet()) {
if (newMetaData.hasIndex(danglingIndex)) {
logger.debug("[{}] no longer dangling (created), removing", danglingIndex);
DanglingIndex removed = danglingIndices.remove(danglingIndex);
removed.future.cancel(false);
}
}
// delete indices that are no longer part of the metadata
try {
for (String indexName : nodeEnv.findAllIndices()) {
// if we have the index on the metadata, don't delete it
if (newMetaData.hasIndex(indexName)) {
continue;
}
if (danglingIndices.containsKey(indexName)) {
// already dangling, continue
continue;
}
IndexMetaData indexMetaData = loadIndex(indexName);
if (indexMetaData != null) {
if (danglingTimeout.millis() == 0) {
logger.info(
"[{}] dangling index, exists on local file system, but not in cluster metadata, timeout set to 0, deleting now",
indexName);
FileSystemUtils.deleteRecursively(nodeEnv.indexLocations(new Index(indexName)));
} else {
logger.info(
"[{}] dangling index, exists on local file system, but not in cluster metadata, scheduling to delete in [{}], auto import to cluster state [{}]",
indexName,
danglingTimeout,
autoImportDangled);
danglingIndices.put(
indexName,
new DanglingIndex(
indexName,
threadPool.schedule(
danglingTimeout,
ThreadPool.Names.SAME,
new RemoveDanglingIndex(indexName))));
}
}
}
} catch (Exception e) {
logger.warn("failed to find dangling indices", e);
}
}
}
if (autoImportDangled.shouldImport() && !danglingIndices.isEmpty()) {
final List<IndexMetaData> dangled = Lists.newArrayList();
for (String indexName : danglingIndices.keySet()) {
IndexMetaData indexMetaData = loadIndex(indexName);
if (indexMetaData == null) {
logger.debug("failed to find state for dangling index [{}]", indexName);
continue;
}
// we might have someone copying over an index, renaming the directory, handle that
if (!indexMetaData.index().equals(indexName)) {
logger.info(
"dangled index directory name is [{}], state name is [{}], renaming to directory name",
indexName,
indexMetaData.index());
indexMetaData =
IndexMetaData.newIndexMetaDataBuilder(indexMetaData).index(indexName).build();
}
if (autoImportDangled == AutoImportDangledState.CLOSED) {
indexMetaData =
IndexMetaData.newIndexMetaDataBuilder(indexMetaData)
.state(IndexMetaData.State.CLOSE)
.build();
}
if (indexMetaData != null) {
dangled.add(indexMetaData);
}
}
IndexMetaData[] dangledIndices = dangled.toArray(new IndexMetaData[dangled.size()]);
try {
allocateDangledIndices.allocateDangled(
dangledIndices,
new LocalAllocateDangledIndices.Listener() {
@Override
public void onResponse(
LocalAllocateDangledIndices.AllocateDangledResponse response) {
logger.trace("allocated dangled");
}
@Override
public void onFailure(Throwable e) {
logger.info("failed to send allocated dangled", e);
}
});
} catch (Exception e) {
logger.warn("failed to send allocate dangled", e);
}
}
}
if (success) {
currentMetaData = newMetaData;
}
}
