private Decision shouldFilter(
ShardRouting shardRouting, RoutingNode node, RoutingAllocation allocation) {
Decision decision = shouldClusterFilter(node, allocation);
if (decision != null) return decision;
decision =
shouldIndexFilter(
allocation.metaData().getIndexSafe(shardRouting.index()), node, allocation);
if (decision != null) return decision;
return allocation.decision(Decision.YES, NAME, "node passes include/exclude/require filters");
}
@Override
protected AsyncShardFetch.FetchResult<
TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData>
fetchData(ShardRouting shard, RoutingAllocation allocation) {
AsyncShardFetch<TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData> fetch =
asyncFetchStore.get(shard.shardId());
if (fetch == null) {
fetch = new InternalAsyncFetch<>(logger, "shard_store", shard.shardId(), storeAction);
asyncFetchStore.put(shard.shardId(), fetch);
}
AsyncShardFetch.FetchResult<TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData>
shardStores =
fetch.fetchData(
allocation.nodes(),
allocation.metaData(),
allocation.getIgnoreNodes(shard.shardId()));
if (shardStores.hasData() == true) {
shardStores.processAllocation(allocation);
}
return shardStores;
}
@Override
protected AsyncShardFetch.FetchResult<
TransportNodesListGatewayStartedShards.NodeGatewayStartedShards>
fetchData(ShardRouting shard, RoutingAllocation allocation) {
AsyncShardFetch<TransportNodesListGatewayStartedShards.NodeGatewayStartedShards> fetch =
asyncFetchStarted.get(shard.shardId());
if (fetch == null) {
fetch = new InternalAsyncFetch<>(logger, "shard_started", shard.shardId(), startedAction);
asyncFetchStarted.put(shard.shardId(), fetch);
}
AsyncShardFetch.FetchResult<TransportNodesListGatewayStartedShards.NodeGatewayStartedShards>
shardState =
fetch.fetchData(
allocation.nodes(),
allocation.metaData(),
allocation.getIgnoreNodes(shard.shardId()));
if (shardState.hasData() == true) {
shardState.processAllocation(allocation);
}
return shardState;
}
@Override
public Decision canAllocate(
ShardRouting shardRouting, RoutingNode node, RoutingAllocation allocation) {
if (shardRouting.unassigned()) {
// only for unassigned - we filter allocation right after the index creation ie. for shard
// shrinking etc. to ensure
// that once it has been allocated post API the replicas can be allocated elsewhere without
// user interaction
// this is a setting that can only be set within the system!
IndexMetaData indexMd = allocation.metaData().getIndexSafe(shardRouting.index());
DiscoveryNodeFilters initialRecoveryFilters = indexMd.getInitialRecoveryFilters();
if (initialRecoveryFilters != null
&& RecoverySource.isInitialRecovery(shardRouting.recoverySource().getType())
&& initialRecoveryFilters.match(node.node()) == false) {
String explanation =
(shardRouting.recoverySource().getType() == RecoverySource.Type.LOCAL_SHARDS)
? "initial allocation of the shrunken index is only allowed on nodes [%s] that hold a copy of every shard in the index"
: "initial allocation of the index is only allowed on nodes [%s]";
return allocation.decision(Decision.NO, NAME, explanation, initialRecoveryFilters);
}
}
return shouldFilter(shardRouting, node, allocation);
}
/**
* 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 boolean processExistingRecoveries(RoutingAllocation allocation) {
boolean changed = false;
MetaData metaData = allocation.metaData();
for (RoutingNodes.RoutingNodesIterator nodes = allocation.routingNodes().nodes();
nodes.hasNext(); ) {
nodes.next();
for (RoutingNodes.RoutingNodeIterator it = nodes.nodeShards(); it.hasNext(); ) {
ShardRouting shard = it.next();
if (shard.primary() == true) {
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...
IndexMetaData indexMetaData = metaData.index(shard.getIndexName());
if (shard.allocatedPostIndexCreate(indexMetaData) == false) {
continue;
}
AsyncShardFetch.FetchResult<TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData>
shardStores = fetchData(shard, allocation);
if (shardStores.hasData() == false) {
logger.trace("{}: fetching new stores for initializing shard", shard);
continue; // still fetching
}
ShardRouting primaryShard = allocation.routingNodes().activePrimary(shard);
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 || primaryStore.allocated() == false) {
// 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);
if (matchingNodes.getNodeWithHighestMatch() != null) {
DiscoveryNode currentNode = allocation.nodes().get(shard.currentNodeId());
DiscoveryNode nodeWithHighestMatch = matchingNodes.getNodeWithHighestMatch();
if (currentNode.equals(nodeWithHighestMatch) == false
&& matchingNodes.isNodeMatchBySyncID(currentNode) == false
&& matchingNodes.isNodeMatchBySyncID(nodeWithHighestMatch) == true) {
// 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
it.moveToUnassigned(
new UnassignedInfo(
UnassignedInfo.Reason.REALLOCATED_REPLICA,
"existing allocation of replica to ["
+ currentNode
+ "] cancelled, sync id match found on node ["
+ nodeWithHighestMatch
+ "]",
null,
allocation.getCurrentNanoTime(),
System.currentTimeMillis()));
changed = true;
}
}
}
}
return changed;
}
public boolean allocateUnassigned(RoutingAllocation allocation) {
boolean changed = false;
final RoutingNodes routingNodes = allocation.routingNodes();
final RoutingNodes.UnassignedShards.UnassignedIterator unassignedIterator =
routingNodes.unassigned().iterator();
MetaData metaData = allocation.metaData();
while (unassignedIterator.hasNext()) {
ShardRouting shard = unassignedIterator.next();
if (shard.primary()) {
continue;
}
// if we are allocating a replica because of index creation, no need to go and find a copy,
// there isn't one...
IndexMetaData indexMetaData = metaData.index(shard.getIndexName());
if (shard.allocatedPostIndexCreate(indexMetaData) == false) {
continue;
}
// pre-check if it can be allocated to any node that currently exists, so we won't list the
// store for it for nothing
if (canBeAllocatedToAtLeastOneNode(shard, allocation) == false) {
logger.trace("{}: ignoring allocation, can't be allocated on any node", shard);
unassignedIterator.removeAndIgnore();
continue;
}
AsyncShardFetch.FetchResult<TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData>
shardStores = fetchData(shard, allocation);
if (shardStores.hasData() == false) {
logger.trace("{}: ignoring allocation, still fetching shard stores", shard);
allocation.setHasPendingAsyncFetch();
unassignedIterator.removeAndIgnore();
continue; // still fetching
}
ShardRouting primaryShard = routingNodes.activePrimary(shard);
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 || primaryStore.allocated() == false) {
// 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",
shard);
continue;
}
MatchingNodes matchingNodes = findMatchingNodes(shard, allocation, primaryStore, shardStores);
if (matchingNodes.getNodeWithHighestMatch() != null) {
RoutingNode nodeWithHighestMatch =
allocation.routingNodes().node(matchingNodes.getNodeWithHighestMatch().id());
// we only check on THROTTLE since we checked before before on NO
Decision decision =
allocation.deciders().canAllocate(shard, nodeWithHighestMatch, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
logger.debug(
"[{}][{}]: throttling allocation [{}] to [{}] in order to reuse its unallocated persistent store",
shard.index(),
shard.id(),
shard,
nodeWithHighestMatch.node());
// we are throttling this, but we have enough to allocate to this node, ignore it for now
unassignedIterator.removeAndIgnore();
} else {
logger.debug(
"[{}][{}]: allocating [{}] to [{}] in order to reuse its unallocated persistent store",
shard.index(),
shard.id(),
shard,
nodeWithHighestMatch.node());
// we found a match
changed = true;
unassignedIterator.initialize(
nodeWithHighestMatch.nodeId(),
null,
allocation
.clusterInfo()
.getShardSize(shard, ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE));
}
} else if (matchingNodes.hasAnyData() == false) {
// if we didn't manage to find *any* data (regardless of matching sizes), check if the
// allocation of the replica shard needs to be delayed
changed |= ignoreUnassignedIfDelayed(unassignedIterator, shard);
}
}
return changed;
}