/**
* Check if the allocation of the replica is to be delayed. Compute the delay and if it is
* delayed, add it to the ignore unassigned list Note: we only care about replica in delayed
* allocation, since if we have an unassigned primary it will anyhow wait to find an existing copy
* of the shard to be allocated Note: the other side of the equation is scheduling a reroute in a
* timely manner, which happens in the RoutingService
*
* <p>PUBLIC FOR TESTS!
*
* @param unassignedIterator iterator over unassigned shards
* @param shard the shard which might be delayed
* @return true iff allocation is delayed for this shard
*/
public boolean ignoreUnassignedIfDelayed(
RoutingNodes.UnassignedShards.UnassignedIterator unassignedIterator, ShardRouting shard) {
// calculate delay and store it in UnassignedInfo to be used by RoutingService
long delay = shard.unassignedInfo().getLastComputedLeftDelayNanos();
if (delay > 0) {
logger.debug(
"[{}][{}]: delaying allocation of [{}] for [{}]",
shard.index(),
shard.id(),
shard,
TimeValue.timeValueNanos(delay));
/**
* mark it as changed, since we want to kick a publishing to schedule future allocation, see
* {@link
* org.elasticsearch.cluster.routing.RoutingService#clusterChanged(ClusterChangedEvent)}).
*/
unassignedIterator.removeAndIgnore();
return true;
}
return false;
}
public boolean allocateUnassigned(RoutingAllocation allocation) {
boolean changed = false;
final RoutingNodes routingNodes = allocation.routingNodes();
final MetaData metaData = routingNodes.metaData();
final RoutingNodes.UnassignedShards.UnassignedIterator unassignedIterator =
routingNodes.unassigned().iterator();
while (unassignedIterator.hasNext()) {
ShardRouting shard = unassignedIterator.next();
if (shard.primary()) {
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
boolean canBeAllocatedToAtLeastOneNode = false;
for (ObjectCursor<DiscoveryNode> cursor : allocation.nodes().dataNodes().values()) {
RoutingNode node = routingNodes.node(cursor.value.id());
if (node == null) {
continue;
}
// if we can't allocate it on a node, ignore it, for example, this handles
// cases for only allocating a replica after a primary
Decision decision = allocation.deciders().canAllocate(shard, node, allocation);
if (decision.type() == Decision.Type.YES) {
canBeAllocatedToAtLeastOneNode = true;
break;
}
}
if (!canBeAllocatedToAtLeastOneNode) {
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);
unassignedIterator.removeAndIgnore();
continue; // still fetching
}
long lastSizeMatched = 0;
DiscoveryNode lastDiscoNodeMatched = null;
RoutingNode lastNodeMatched = null;
boolean hasReplicaData = false;
IndexMetaData indexMetaData = metaData.index(shard.getIndex());
for (Map.Entry<DiscoveryNode, TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData>
nodeStoreEntry : shardStores.getData().entrySet()) {
DiscoveryNode discoNode = nodeStoreEntry.getKey();
TransportNodesListShardStoreMetaData.StoreFilesMetaData storeFilesMetaData =
nodeStoreEntry.getValue().storeFilesMetaData();
logger.trace("{}: checking node [{}]", shard, discoNode);
if (storeFilesMetaData == null) {
// already allocated on that node...
continue;
}
RoutingNode node = routingNodes.node(discoNode.id());
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);
if (decision.type() == Decision.Type.NO) {
continue;
}
// if it is already allocated, we can't assign to it...
if (storeFilesMetaData.allocated()) {
continue;
}
if (!shard.primary()) {
hasReplicaData |= storeFilesMetaData.iterator().hasNext();
ShardRouting primaryShard = routingNodes.activePrimary(shard);
if (primaryShard != null) {
assert primaryShard.active();
DiscoveryNode primaryNode = allocation.nodes().get(primaryShard.currentNodeId());
if (primaryNode != null) {
TransportNodesListShardStoreMetaData.NodeStoreFilesMetaData primaryNodeFilesStore =
shardStores.getData().get(primaryNode);
if (primaryNodeFilesStore != null) {
TransportNodesListShardStoreMetaData.StoreFilesMetaData primaryNodeStore =
primaryNodeFilesStore.storeFilesMetaData();
if (primaryNodeStore != null && primaryNodeStore.allocated()) {
long sizeMatched = 0;
String primarySyncId = primaryNodeStore.syncId();
String replicaSyncId = storeFilesMetaData.syncId();
// see if we have a sync id we can make use of
if (replicaSyncId != null && replicaSyncId.equals(primarySyncId)) {
logger.trace(
"{}: node [{}] has same sync id {} as primary",
shard,
discoNode.name(),
replicaSyncId);
lastNodeMatched = node;
lastSizeMatched = Long.MAX_VALUE;
lastDiscoNodeMatched = discoNode;
} else {
for (StoreFileMetaData storeFileMetaData : storeFilesMetaData) {
String metaDataFileName = storeFileMetaData.name();
if (primaryNodeStore.fileExists(metaDataFileName)
&& primaryNodeStore.file(metaDataFileName).isSame(storeFileMetaData)) {
sizeMatched += storeFileMetaData.length();
}
}
logger.trace(
"{}: node [{}] has [{}/{}] bytes of re-usable data",
shard,
discoNode.name(),
new ByteSizeValue(sizeMatched),
sizeMatched);
if (sizeMatched > lastSizeMatched) {
lastSizeMatched = sizeMatched;
lastDiscoNodeMatched = discoNode;
lastNodeMatched = node;
}
}
}
}
}
}
}
}
if (lastNodeMatched != null) {
// we only check on THROTTLE since we checked before before on NO
Decision decision = allocation.deciders().canAllocate(shard, lastNodeMatched, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: throttling allocation [{}] to [{}] in order to reuse its unallocated persistent store with total_size [{}]",
shard.index(),
shard.id(),
shard,
lastDiscoNodeMatched,
new ByteSizeValue(lastSizeMatched));
}
// we are throttling this, but we have enough to allocate to this node, ignore it for now
unassignedIterator.removeAndIgnore();
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: allocating [{}] to [{}] in order to reuse its unallocated persistent store with total_size [{}]",
shard.index(),
shard.id(),
shard,
lastDiscoNodeMatched,
new ByteSizeValue(lastSizeMatched));
}
// we found a match
changed = true;
unassignedIterator.initialize(lastNodeMatched.nodeId());
}
} else if (hasReplicaData == 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, and if so, add it to the ignore unassigned list
// note: we only care about replica in delayed allocation, since if we have an unassigned
// primary it
// will anyhow wait to find an existing copy of the shard to be allocated
// note: the other side of the equation is scheduling a reroute in a timely manner, which
// happens in the RoutingService
long delay =
shard
.unassignedInfo()
.getDelayAllocationExpirationIn(settings, indexMetaData.getSettings());
if (delay > 0) {
logger.debug(
"[{}][{}]: delaying allocation of [{}] for [{}]",
shard.index(),
shard.id(),
shard,
TimeValue.timeValueMillis(delay));
/**
* mark it as changed, since we want to kick a publishing to schedule future allocation,
* see {@link
* org.elasticsearch.cluster.routing.RoutingService#clusterChanged(ClusterChangedEvent)}).
*/
changed = true;
unassignedIterator.removeAndIgnore();
}
}
}
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;
}
@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
final DiscoveryNode discoNode = allocation.nodes().resolveNode(node);
final RoutingNodes routingNodes = allocation.routingNodes();
ShardRouting shardRouting = null;
for (ShardRouting routing : routingNodes.unassigned()) {
if (routing.shardId().equals(shardId)) {
// prefer primaries first to allocate
if (shardRouting == null || routing.primary()) {
shardRouting = routing;
}
}
}
if (shardRouting == null) {
if (explain) {
return new RerouteExplanation(
this,
allocation.decision(
Decision.NO,
"allocate_allocation_command",
"failed to find " + shardId + " on the list of unassigned shards"));
}
throw new IllegalArgumentException(
"[allocate] failed to find " + shardId + " on the list of unassigned shards");
}
if (shardRouting.primary() && !allowPrimary) {
if (explain) {
return new RerouteExplanation(
this,
allocation.decision(
Decision.NO,
"allocate_allocation_command",
"trying to allocate a primary shard " + shardId + ", which is disabled"));
}
throw new IllegalArgumentException(
"[allocate] trying to allocate a primary shard " + shardId + ", which is disabled");
}
RoutingNode routingNode = routingNodes.node(discoNode.id());
if (routingNode == null) {
if (!discoNode.dataNode()) {
if (explain) {
return new RerouteExplanation(
this,
allocation.decision(
Decision.NO,
"allocate_allocation_command",
"Allocation can only be done on data nodes, not [" + node + "]"));
}
throw new IllegalArgumentException(
"Allocation can only be done on data nodes, not [" + node + "]");
} else {
if (explain) {
return new RerouteExplanation(
this,
allocation.decision(
Decision.NO,
"allocate_allocation_command",
"Could not find [" + node + "] among the routing nodes"));
}
throw new IllegalStateException("Could not find [" + node + "] among the routing nodes");
}
}
Decision decision = allocation.deciders().canAllocate(shardRouting, routingNode, allocation);
if (decision.type() == Decision.Type.NO) {
if (explain) {
return new RerouteExplanation(this, decision);
}
throw new IllegalArgumentException(
"[allocate] allocation of "
+ shardId
+ " on node "
+ discoNode
+ " is not allowed, reason: "
+ decision);
}
// go over and remove it from the unassigned
for (RoutingNodes.UnassignedShards.UnassignedIterator it = routingNodes.unassigned().iterator();
it.hasNext(); ) {
if (it.next() != shardRouting) {
continue;
}
it.initialize(routingNode.nodeId());
if (shardRouting.primary()) {
// we need to clear the post allocation flag, since its an explicit allocation of the
// primary shard
// and we want to force allocate it (and create a new index for it)
routingNodes.addClearPostAllocationFlag(shardRouting.shardId());
}
break;
}
return new RerouteExplanation(this, decision);
}
