use of org.elasticsearch.indices.store.TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData in project elasticsearch by elastic.
the class ReplicaShardAllocator method findStore.
/**
* Finds the store for the assigned shard in the fetched data, returns null if none is found.
*/
private TransportNodesListShardStoreMetaData.StoreFilesMetaData findStore(ShardRouting shard, RoutingAllocation allocation, AsyncShardFetch.FetchResult<NodeStoreFilesMetaData> data) {
assert shard.currentNodeId() != null;
DiscoveryNode primaryNode = allocation.nodes().get(shard.currentNodeId());
if (primaryNode == null) {
return null;
}
NodeStoreFilesMetaData primaryNodeFilesStore = data.getData().get(primaryNode);
if (primaryNodeFilesStore == null) {
return null;
}
return primaryNodeFilesStore.storeFilesMetaData();
}
use of org.elasticsearch.indices.store.TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData in project elasticsearch by elastic.
the class ReplicaShardAllocator method makeAllocationDecision.
@Override
public AllocateUnassignedDecision makeAllocationDecision(final ShardRouting unassignedShard, final RoutingAllocation allocation, final Logger logger) {
if (isResponsibleFor(unassignedShard) == false) {
// this allocator is not responsible for deciding on this shard
return AllocateUnassignedDecision.NOT_TAKEN;
}
final RoutingNodes routingNodes = allocation.routingNodes();
final boolean explain = allocation.debugDecision();
// pre-check if it can be allocated to any node that currently exists, so we won't list the store for it for nothing
Tuple<Decision, Map<String, NodeAllocationResult>> result = canBeAllocatedToAtLeastOneNode(unassignedShard, allocation);
Decision allocateDecision = result.v1();
if (allocateDecision.type() != Decision.Type.YES && (explain == false || hasInitiatedFetching(unassignedShard) == false)) {
// only return early if we are not in explain mode, or we are in explain mode but we have not
// yet attempted to fetch any shard data
logger.trace("{}: ignoring allocation, can't be allocated on any node", unassignedShard);
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), result.v2() != null ? new ArrayList<>(result.v2().values()) : null);
}
AsyncShardFetch.FetchResult<NodeStoreFilesMetaData> shardStores = fetchData(unassignedShard, allocation);
if (shardStores.hasData() == false) {
logger.trace("{}: ignoring allocation, still fetching shard stores", unassignedShard);
allocation.setHasPendingAsyncFetch();
List<NodeAllocationResult> nodeDecisions = null;
if (explain) {
nodeDecisions = buildDecisionsForAllNodes(unassignedShard, allocation);
}
return AllocateUnassignedDecision.no(AllocationStatus.FETCHING_SHARD_DATA, nodeDecisions);
}
ShardRouting primaryShard = routingNodes.activePrimary(unassignedShard.shardId());
if (primaryShard == null) {
assert explain : "primary should only be null here if we are in explain mode, so we didn't " + "exit early when canBeAllocatedToAtLeastOneNode didn't return a YES decision";
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), new ArrayList<>(result.v2().values()));
}
TransportNodesListShardStoreMetaData.StoreFilesMetaData primaryStore = findStore(primaryShard, allocation, shardStores);
if (primaryStore == null) {
// if we can't find the primary data, it is probably because the primary shard is corrupted (and listing failed)
// we want to let the replica be allocated in order to expose the actual problem with the primary that the replica
// will try and recover from
// Note, this is the existing behavior, as exposed in running CorruptFileTest#testNoPrimaryData
logger.trace("{}: no primary shard store found or allocated, letting actual allocation figure it out", unassignedShard);
return AllocateUnassignedDecision.NOT_TAKEN;
}
MatchingNodes matchingNodes = findMatchingNodes(unassignedShard, allocation, primaryStore, shardStores, explain);
assert explain == false || matchingNodes.nodeDecisions != null : "in explain mode, we must have individual node decisions";
List<NodeAllocationResult> nodeDecisions = augmentExplanationsWithStoreInfo(result.v2(), matchingNodes.nodeDecisions);
if (allocateDecision.type() != Decision.Type.YES) {
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), nodeDecisions);
} else if (matchingNodes.getNodeWithHighestMatch() != null) {
RoutingNode nodeWithHighestMatch = allocation.routingNodes().node(matchingNodes.getNodeWithHighestMatch().getId());
// we only check on THROTTLE since we checked before before on NO
Decision decision = allocation.deciders().canAllocate(unassignedShard, nodeWithHighestMatch, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
logger.debug("[{}][{}]: throttling allocation [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
// we are throttling this, as we have enough other shards to allocate to this node, so ignore it for now
return AllocateUnassignedDecision.throttle(nodeDecisions);
} else {
logger.debug("[{}][{}]: allocating [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
// we found a match
return AllocateUnassignedDecision.yes(nodeWithHighestMatch.node(), null, nodeDecisions, true);
}
} else if (matchingNodes.hasAnyData() == false && unassignedShard.unassignedInfo().isDelayed()) {
// if we didn't manage to find *any* data (regardless of matching sizes), and the replica is
// unassigned due to a node leaving, so we delay allocation of this replica to see if the
// node with the shard copy will rejoin so we can re-use the copy it has
logger.debug("{}: allocation of [{}] is delayed", unassignedShard.shardId(), unassignedShard);
long remainingDelayMillis = 0L;
long totalDelayMillis = 0L;
if (explain) {
UnassignedInfo unassignedInfo = unassignedShard.unassignedInfo();
MetaData metadata = allocation.metaData();
IndexMetaData indexMetaData = metadata.index(unassignedShard.index());
totalDelayMillis = INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING.get(indexMetaData.getSettings()).getMillis();
long remainingDelayNanos = unassignedInfo.getRemainingDelay(System.nanoTime(), indexMetaData.getSettings());
remainingDelayMillis = TimeValue.timeValueNanos(remainingDelayNanos).millis();
}
return AllocateUnassignedDecision.delayed(remainingDelayMillis, totalDelayMillis, nodeDecisions);
}
return AllocateUnassignedDecision.NOT_TAKEN;
}
use of org.elasticsearch.indices.store.TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData in project elasticsearch by elastic.
the class ReplicaShardAllocator method processExistingRecoveries.
/**
* Process existing recoveries of replicas and see if we need to cancel them if we find a better
* match. Today, a better match is one that has full sync id match compared to not having one in
* the previous recovery.
*/
public void processExistingRecoveries(RoutingAllocation allocation) {
MetaData metaData = allocation.metaData();
RoutingNodes routingNodes = allocation.routingNodes();
List<Runnable> shardCancellationActions = new ArrayList<>();
for (RoutingNode routingNode : routingNodes) {
for (ShardRouting shard : routingNode) {
if (shard.primary()) {
continue;
}
if (shard.initializing() == false) {
continue;
}
if (shard.relocatingNodeId() != null) {
continue;
}
// if we are allocating a replica because of index creation, no need to go and find a copy, there isn't one...
if (shard.unassignedInfo() != null && shard.unassignedInfo().getReason() == UnassignedInfo.Reason.INDEX_CREATED) {
continue;
}
AsyncShardFetch.FetchResult<NodeStoreFilesMetaData> shardStores = fetchData(shard, allocation);
if (shardStores.hasData() == false) {
logger.trace("{}: fetching new stores for initializing shard", shard);
// still fetching
continue;
}
ShardRouting primaryShard = allocation.routingNodes().activePrimary(shard.shardId());
assert primaryShard != null : "the replica shard can be allocated on at least one node, so there must be an active primary";
TransportNodesListShardStoreMetaData.StoreFilesMetaData primaryStore = findStore(primaryShard, allocation, shardStores);
if (primaryStore == null) {
// if we can't find the primary data, it is probably because the primary shard is corrupted (and listing failed)
// just let the recovery find it out, no need to do anything about it for the initializing shard
logger.trace("{}: no primary shard store found or allocated, letting actual allocation figure it out", shard);
continue;
}
MatchingNodes matchingNodes = findMatchingNodes(shard, allocation, primaryStore, shardStores, false);
if (matchingNodes.getNodeWithHighestMatch() != null) {
DiscoveryNode currentNode = allocation.nodes().get(shard.currentNodeId());
DiscoveryNode nodeWithHighestMatch = matchingNodes.getNodeWithHighestMatch();
// current node will not be in matchingNodes as it is filtered away by SameShardAllocationDecider
final String currentSyncId;
if (shardStores.getData().containsKey(currentNode)) {
currentSyncId = shardStores.getData().get(currentNode).storeFilesMetaData().syncId();
} else {
currentSyncId = null;
}
if (currentNode.equals(nodeWithHighestMatch) == false && Objects.equals(currentSyncId, primaryStore.syncId()) == false && matchingNodes.isNodeMatchBySyncID(nodeWithHighestMatch)) {
// we found a better match that has a full sync id match, the existing allocation is not fully synced
// so we found a better one, cancel this one
logger.debug("cancelling allocation of replica on [{}], sync id match found on node [{}]", currentNode, nodeWithHighestMatch);
UnassignedInfo unassignedInfo = new UnassignedInfo(UnassignedInfo.Reason.REALLOCATED_REPLICA, "existing allocation of replica to [" + currentNode + "] cancelled, sync id match found on node [" + nodeWithHighestMatch + "]", null, 0, allocation.getCurrentNanoTime(), System.currentTimeMillis(), false, UnassignedInfo.AllocationStatus.NO_ATTEMPT);
// don't cancel shard in the loop as it will cause a ConcurrentModificationException
shardCancellationActions.add(() -> routingNodes.failShard(logger, shard, unassignedInfo, metaData.getIndexSafe(shard.index()), allocation.changes()));
}
}
}
}
for (Runnable action : shardCancellationActions) {
action.run();
}
}
use of org.elasticsearch.indices.store.TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData in project elasticsearch by elastic.
the class ReplicaShardAllocator method findMatchingNodes.
private MatchingNodes findMatchingNodes(ShardRouting shard, RoutingAllocation allocation, TransportNodesListShardStoreMetaData.StoreFilesMetaData primaryStore, AsyncShardFetch.FetchResult<NodeStoreFilesMetaData> data, boolean explain) {
ObjectLongMap<DiscoveryNode> nodesToSize = new ObjectLongHashMap<>();
Map<String, NodeAllocationResult> nodeDecisions = explain ? new HashMap<>() : null;
for (Map.Entry<DiscoveryNode, NodeStoreFilesMetaData> nodeStoreEntry : data.getData().entrySet()) {
DiscoveryNode discoNode = nodeStoreEntry.getKey();
TransportNodesListShardStoreMetaData.StoreFilesMetaData storeFilesMetaData = nodeStoreEntry.getValue().storeFilesMetaData();
// we don't have any files at all, it is an empty index
if (storeFilesMetaData.isEmpty()) {
continue;
}
RoutingNode node = allocation.routingNodes().node(discoNode.getId());
if (node == null) {
continue;
}
// check if we can allocate on that node...
// we only check for NO, since if this node is THROTTLING and it has enough "same data"
// then we will try and assign it next time
Decision decision = allocation.deciders().canAllocate(shard, node, allocation);
long matchingBytes = -1;
if (explain) {
matchingBytes = computeMatchingBytes(primaryStore, storeFilesMetaData);
ShardStoreInfo shardStoreInfo = new ShardStoreInfo(matchingBytes);
nodeDecisions.put(node.nodeId(), new NodeAllocationResult(discoNode, shardStoreInfo, decision));
}
if (decision.type() == Decision.Type.NO) {
continue;
}
if (matchingBytes < 0) {
matchingBytes = computeMatchingBytes(primaryStore, storeFilesMetaData);
}
nodesToSize.put(discoNode, matchingBytes);
if (logger.isTraceEnabled()) {
if (matchingBytes == Long.MAX_VALUE) {
logger.trace("{}: node [{}] has same sync id {} as primary", shard, discoNode.getName(), storeFilesMetaData.syncId());
} else {
logger.trace("{}: node [{}] has [{}/{}] bytes of re-usable data", shard, discoNode.getName(), new ByteSizeValue(matchingBytes), matchingBytes);
}
}
}
return new MatchingNodes(nodesToSize, nodeDecisions);
}
use of org.elasticsearch.indices.store.TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData in project crate by crate.
the class ReplicaShardAllocator method makeAllocationDecision.
@Override
public AllocateUnassignedDecision makeAllocationDecision(final ShardRouting unassignedShard, final RoutingAllocation allocation, final Logger logger) {
if (isResponsibleFor(unassignedShard) == false) {
// this allocator is not responsible for deciding on this shard
return AllocateUnassignedDecision.NOT_TAKEN;
}
final RoutingNodes routingNodes = allocation.routingNodes();
final boolean explain = allocation.debugDecision();
// pre-check if it can be allocated to any node that currently exists, so we won't list the store for it for nothing
Tuple<Decision, Map<String, NodeAllocationResult>> result = canBeAllocatedToAtLeastOneNode(unassignedShard, allocation);
Decision allocateDecision = result.v1();
if (allocateDecision.type() != Decision.Type.YES && (explain == false || hasInitiatedFetching(unassignedShard) == false)) {
// only return early if we are not in explain mode, or we are in explain mode but we have not
// yet attempted to fetch any shard data
logger.trace("{}: ignoring allocation, can't be allocated on any node", unassignedShard);
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), result.v2() != null ? new ArrayList<>(result.v2().values()) : null);
}
AsyncShardFetch.FetchResult<NodeStoreFilesMetadata> shardStores = fetchData(unassignedShard, allocation);
if (shardStores.hasData() == false) {
logger.trace("{}: ignoring allocation, still fetching shard stores", unassignedShard);
allocation.setHasPendingAsyncFetch();
List<NodeAllocationResult> nodeDecisions = null;
if (explain) {
nodeDecisions = buildDecisionsForAllNodes(unassignedShard, allocation);
}
return AllocateUnassignedDecision.no(AllocationStatus.FETCHING_SHARD_DATA, nodeDecisions);
}
ShardRouting primaryShard = routingNodes.activePrimary(unassignedShard.shardId());
if (primaryShard == null) {
assert explain : "primary should only be null here if we are in explain mode, so we didn't " + "exit early when canBeAllocatedToAtLeastOneNode didn't return a YES decision";
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), new ArrayList<>(result.v2().values()));
}
assert primaryShard.currentNodeId() != null;
final DiscoveryNode primaryNode = allocation.nodes().get(primaryShard.currentNodeId());
final TransportNodesListShardStoreMetadata.StoreFilesMetadata primaryStore = findStore(primaryNode, shardStores);
if (primaryStore == null) {
// if we can't find the primary data, it is probably because the primary shard is corrupted (and listing failed)
// we want to let the replica be allocated in order to expose the actual problem with the primary that the replica
// will try and recover from
// Note, this is the existing behavior, as exposed in running CorruptFileTest#testNoPrimaryData
logger.trace("{}: no primary shard store found or allocated, letting actual allocation figure it out", unassignedShard);
return AllocateUnassignedDecision.NOT_TAKEN;
}
MatchingNodes matchingNodes = findMatchingNodes(unassignedShard, allocation, false, primaryNode, primaryStore, shardStores, explain);
assert explain == false || matchingNodes.nodeDecisions != null : "in explain mode, we must have individual node decisions";
List<NodeAllocationResult> nodeDecisions = augmentExplanationsWithStoreInfo(result.v2(), matchingNodes.nodeDecisions);
if (allocateDecision.type() != Decision.Type.YES) {
return AllocateUnassignedDecision.no(UnassignedInfo.AllocationStatus.fromDecision(allocateDecision.type()), nodeDecisions);
} else if (matchingNodes.getNodeWithHighestMatch() != null) {
RoutingNode nodeWithHighestMatch = allocation.routingNodes().node(matchingNodes.getNodeWithHighestMatch().getId());
// we only check on THROTTLE since we checked before before on NO
Decision decision = allocation.deciders().canAllocate(unassignedShard, nodeWithHighestMatch, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
logger.debug("[{}][{}]: throttling allocation [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
// we are throttling this, as we have enough other shards to allocate to this node, so ignore it for now
return AllocateUnassignedDecision.throttle(nodeDecisions);
} else {
logger.debug("[{}][{}]: allocating [{}] to [{}] in order to reuse its unallocated persistent store", unassignedShard.index(), unassignedShard.id(), unassignedShard, nodeWithHighestMatch.node());
// we found a match
return AllocateUnassignedDecision.yes(nodeWithHighestMatch.node(), null, nodeDecisions, true);
}
} else if (matchingNodes.hasAnyData() == false && unassignedShard.unassignedInfo().isDelayed()) {
// if we didn't manage to find *any* data (regardless of matching sizes), and the replica is
// unassigned due to a node leaving, so we delay allocation of this replica to see if the
// node with the shard copy will rejoin so we can re-use the copy it has
logger.debug("{}: allocation of [{}] is delayed", unassignedShard.shardId(), unassignedShard);
long remainingDelayMillis = 0L;
long totalDelayMillis = 0L;
if (explain) {
UnassignedInfo unassignedInfo = unassignedShard.unassignedInfo();
Metadata metadata = allocation.metadata();
IndexMetadata indexMetadata = metadata.index(unassignedShard.index());
totalDelayMillis = INDEX_DELAYED_NODE_LEFT_TIMEOUT_SETTING.get(indexMetadata.getSettings()).getMillis();
long remainingDelayNanos = unassignedInfo.getRemainingDelay(System.nanoTime(), indexMetadata.getSettings());
remainingDelayMillis = TimeValue.timeValueNanos(remainingDelayNanos).millis();
}
return AllocateUnassignedDecision.delayed(remainingDelayMillis, totalDelayMillis, nodeDecisions);
}
return AllocateUnassignedDecision.NOT_TAKEN;
}
Aggregations