/** Initializes an index, to be restored from snapshot */
private Builder initializeAsRestore(
IndexMetaData indexMetaData, RestoreSource restoreSource, boolean asNew) {
if (!shards.isEmpty()) {
throw new ElasticSearchIllegalStateException(
"trying to initialize an index with fresh shards, but already has shards created");
}
for (int shardId = 0; shardId < indexMetaData.numberOfShards(); shardId++) {
IndexShardRoutingTable.Builder indexShardRoutingBuilder =
new IndexShardRoutingTable.Builder(
new ShardId(indexMetaData.index(), shardId), asNew ? false : true);
for (int i = 0; i <= indexMetaData.numberOfReplicas(); i++) {
indexShardRoutingBuilder.addShard(
new ImmutableShardRouting(
index,
shardId,
null,
null,
i == 0 ? restoreSource : null,
i == 0,
ShardRoutingState.UNASSIGNED,
0));
}
shards.put(shardId, indexShardRoutingBuilder.build());
}
return this;
}
/** validate based on a meta data, returning failures found */
public List<String> validate(IndexMetaData indexMetaData) {
ArrayList<String> failures = new ArrayList<String>();
// check the number of shards
if (indexMetaData.numberOfShards() != shards().size()) {
Set<Integer> expected = Sets.newHashSet();
for (int i = 0; i < indexMetaData.numberOfShards(); i++) {
expected.add(i);
}
for (IndexShardRoutingTable indexShardRoutingTable : this) {
expected.remove(indexShardRoutingTable.shardId().id());
}
failures.add("Wrong number of shards in routing table, missing: " + expected);
}
// check the replicas
for (IndexShardRoutingTable indexShardRoutingTable : this) {
int routingNumberOfReplicas = indexShardRoutingTable.size() - 1;
if (routingNumberOfReplicas != indexMetaData.numberOfReplicas()) {
failures.add(
"Shard ["
+ indexShardRoutingTable.shardId().id()
+ "] routing table has wrong number of replicas, expected ["
+ indexMetaData.numberOfReplicas()
+ "], got ["
+ routingNumberOfReplicas
+ "]");
}
for (ShardRouting shardRouting : indexShardRoutingTable) {
if (!shardRouting.index().equals(index())) {
failures.add(
"shard routing has an index ["
+ shardRouting.index()
+ "] that is different than the routing table");
}
}
}
return failures;
}
public boolean allocateUnassigned(RoutingAllocation allocation) {
boolean changed = false;
DiscoveryNodes nodes = allocation.nodes();
RoutingNodes routingNodes = allocation.routingNodes();
// First, handle primaries, they must find a place to be allocated on here
final MetaData metaData = routingNodes.metaData();
RoutingNodes.UnassignedShards unassigned = routingNodes.unassigned();
unassigned.sort(
new PriorityComparator() {
@Override
protected Settings getIndexSettings(String index) {
IndexMetaData indexMetaData = metaData.index(index);
return indexMetaData.getSettings();
}
}); // sort for priority ordering
Iterator<ShardRouting> unassignedIterator = unassigned.iterator();
while (unassignedIterator.hasNext()) {
ShardRouting shard = unassignedIterator.next();
if (!shard.primary()) {
continue;
}
// this is an API allocation, ignore since we know there is no data...
if (!routingNodes
.routingTable()
.index(shard.index())
.shard(shard.id())
.primaryAllocatedPostApi()) {
continue;
}
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(nodes, metaData, allocation.getIgnoreNodes(shard.shardId()));
if (shardState.hasData() == false) {
logger.trace("{}: ignoring allocation, still fetching shard started state", shard);
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
continue;
}
shardState.processAllocation(allocation);
IndexMetaData indexMetaData = metaData.index(shard.getIndex());
/**
* Build a map of DiscoveryNodes to shard state number for the given shard. A state of -1
* means the shard does not exist on the node, where any shard state >= 0 is the state version
* of the shard on that node's disk.
*
* <p>A shard on shared storage will return at least shard state 0 for all nodes, indicating
* that the shard can be allocated to any node.
*/
ObjectLongHashMap<DiscoveryNode> nodesState = new ObjectLongHashMap<>();
for (TransportNodesListGatewayStartedShards.NodeGatewayStartedShards nodeShardState :
shardState.getData().values()) {
long version = nodeShardState.version();
// -1 version means it does not exists, which is what the API returns, and what we expect to
logger.trace(
"[{}] on node [{}] has version [{}] of shard",
shard,
nodeShardState.getNode(),
version);
nodesState.put(nodeShardState.getNode(), version);
}
int numberOfAllocationsFound = 0;
long highestVersion = -1;
final Map<DiscoveryNode, Long> nodesWithVersion = Maps.newHashMap();
assert !nodesState.containsKey(null);
final Object[] keys = nodesState.keys;
final long[] values = nodesState.values;
Settings idxSettings = indexMetaData.settings();
for (int i = 0; i < keys.length; i++) {
if (keys[i] == null) {
continue;
}
DiscoveryNode node = (DiscoveryNode) keys[i];
long version = values[i];
// since we don't check in NO allocation, we need to double check here
if (allocation.shouldIgnoreShardForNode(shard.shardId(), node.id())) {
continue;
}
if (recoverOnAnyNode(idxSettings)) {
numberOfAllocationsFound++;
if (version > highestVersion) {
highestVersion = version;
}
// We always put the node without clearing the map
nodesWithVersion.put(node, version);
} else if (version != -1) {
numberOfAllocationsFound++;
// If we've found a new "best" candidate, clear the
// current candidates and add it
if (version > highestVersion) {
highestVersion = version;
nodesWithVersion.clear();
nodesWithVersion.put(node, version);
} else if (version == highestVersion) {
// If the candidate is the same, add it to the
// list, but keep the current candidate
nodesWithVersion.put(node, version);
}
}
}
// Now that we have a map of nodes to versions along with the
// number of allocations found (and not ignored), we need to sort
// it so the node with the highest version is at the beginning
List<DiscoveryNode> nodesWithHighestVersion = Lists.newArrayList();
nodesWithHighestVersion.addAll(nodesWithVersion.keySet());
CollectionUtil.timSort(
nodesWithHighestVersion,
new Comparator<DiscoveryNode>() {
@Override
public int compare(DiscoveryNode o1, DiscoveryNode o2) {
return Long.compare(nodesWithVersion.get(o2), nodesWithVersion.get(o1));
}
});
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}] found {} allocations of {}, highest version: [{}]",
shard.index(),
shard.id(),
numberOfAllocationsFound,
shard,
highestVersion);
}
if (logger.isTraceEnabled()) {
StringBuilder sb = new StringBuilder("[");
for (DiscoveryNode n : nodesWithHighestVersion) {
sb.append("[");
sb.append(n.getName());
sb.append("]");
sb.append(" -> ");
sb.append(nodesWithVersion.get(n));
sb.append(", ");
}
sb.append("]");
logger.trace("{} candidates for allocation: {}", shard, sb.toString());
}
// check if the counts meets the minimum set
int requiredAllocation = 1;
// if we restore from a repository one copy is more then enough
if (shard.restoreSource() == null) {
try {
String initialShards =
indexMetaData
.settings()
.get(
INDEX_RECOVERY_INITIAL_SHARDS,
settings.get(INDEX_RECOVERY_INITIAL_SHARDS, this.initialShards));
if ("quorum".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 1) {
requiredAllocation = ((1 + indexMetaData.numberOfReplicas()) / 2) + 1;
}
} else if ("quorum-1".equals(initialShards) || "half".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 2) {
requiredAllocation = ((1 + indexMetaData.numberOfReplicas()) / 2);
}
} else if ("one".equals(initialShards)) {
requiredAllocation = 1;
} else if ("full".equals(initialShards) || "all".equals(initialShards)) {
requiredAllocation = indexMetaData.numberOfReplicas() + 1;
} else if ("full-1".equals(initialShards) || "all-1".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 1) {
requiredAllocation = indexMetaData.numberOfReplicas();
}
} else {
requiredAllocation = Integer.parseInt(initialShards);
}
} catch (Exception e) {
logger.warn(
"[{}][{}] failed to derived initial_shards from value {}, ignore allocation for {}",
shard.index(),
shard.id(),
initialShards,
shard);
}
}
// not enough found for this shard, continue...
if (numberOfAllocationsFound < requiredAllocation) {
// if we are restoring this shard we still can allocate
if (shard.restoreSource() == null) {
// we can't really allocate, so ignore it and continue
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: not allocating, number_of_allocated_shards_found [{}], required_number [{}]",
shard.index(),
shard.id(),
numberOfAllocationsFound,
requiredAllocation);
}
} else if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: missing local data, will restore from [{}]",
shard.index(),
shard.id(),
shard.restoreSource());
}
continue;
}
Set<DiscoveryNode> throttledNodes = Sets.newHashSet();
Set<DiscoveryNode> noNodes = Sets.newHashSet();
for (DiscoveryNode discoNode : nodesWithHighestVersion) {
RoutingNode node = routingNodes.node(discoNode.id());
if (node == null) {
continue;
}
Decision decision = allocation.deciders().canAllocate(shard, node, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
throttledNodes.add(discoNode);
} else if (decision.type() == Decision.Type.NO) {
noNodes.add(discoNode);
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: allocating [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
discoNode);
}
// we found a match
changed = true;
// make sure we create one with the version from the recovered state
routingNodes.initialize(new ShardRouting(shard, highestVersion), node.nodeId());
unassignedIterator.remove();
// found a node, so no throttling, no "no", and break out of the loop
throttledNodes.clear();
noNodes.clear();
break;
}
}
if (throttledNodes.isEmpty()) {
// if we have a node that we "can't" allocate to, force allocation, since this is our master
// data!
if (!noNodes.isEmpty()) {
DiscoveryNode discoNode = noNodes.iterator().next();
RoutingNode node = routingNodes.node(discoNode.id());
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: forcing allocating [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
discoNode);
}
// we found a match
changed = true;
// make sure we create one with the version from the recovered state
routingNodes.initialize(new ShardRouting(shard, highestVersion), node.nodeId());
unassignedIterator.remove();
}
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: throttling allocation [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
throttledNodes);
}
// we are throttling this, but we have enough to allocate to this node, ignore it for now
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
}
}
if (!routingNodes.hasUnassigned()) {
return changed;
}
// Now, handle replicas, try to assign them to nodes that are similar to the one the primary was
// allocated on
unassignedIterator = 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 : 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.remove();
routingNodes.ignoredUnassigned().add(shard);
continue;
}
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(nodes, metaData, allocation.getIgnoreNodes(shard.shardId()));
if (shardStores.hasData() == false) {
logger.trace("{}: ignoring allocation, still fetching shard stores", shard);
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
continue; // still fetching
}
shardStores.processAllocation(allocation);
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 = 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.remove();
routingNodes.ignoredUnassigned().add(shard);
} 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;
routingNodes.initialize(shard, lastNodeMatched.nodeId());
unassignedIterator.remove();
}
} 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.remove();
routingNodes.ignoredUnassigned().add(shard);
}
}
}
return changed;
}
@Test
public void testSimpleJsonFromAndTo() throws IOException {
MetaData metaData =
newMetaDataBuilder()
.maxNumberOfShardsPerNode(2)
.put(newIndexMetaDataBuilder("test1").numberOfShards(1).numberOfReplicas(2))
.put(
newIndexMetaDataBuilder("test2")
.settings(settingsBuilder().put("setting1", "value1").put("setting2", "value2"))
.numberOfShards(2)
.numberOfReplicas(3))
.put(
newIndexMetaDataBuilder("test3")
.numberOfShards(1)
.numberOfReplicas(2)
.putMapping("mapping1", MAPPING_SOURCE1))
.put(
newIndexMetaDataBuilder("test4")
.settings(settingsBuilder().put("setting1", "value1").put("setting2", "value2"))
.numberOfShards(1)
.numberOfReplicas(2)
.putMapping("mapping1", MAPPING_SOURCE1)
.putMapping("mapping2", MAPPING_SOURCE2))
.build();
String metaDataSource = MetaData.Builder.toJson(metaData);
System.out.println("ToJson: " + metaDataSource);
MetaData parsedMetaData =
MetaData.Builder.fromJson(
Jackson.defaultJsonFactory().createJsonParser(metaDataSource), null);
assertThat(parsedMetaData.maxNumberOfShardsPerNode(), equalTo(2));
IndexMetaData indexMetaData = metaData.index("test1");
assertThat(indexMetaData.numberOfShards(), equalTo(1));
assertThat(indexMetaData.numberOfReplicas(), equalTo(2));
assertThat(indexMetaData.settings().getAsMap().size(), equalTo(2));
assertThat(indexMetaData.mappings().size(), equalTo(0));
indexMetaData = metaData.index("test2");
assertThat(indexMetaData.numberOfShards(), equalTo(2));
assertThat(indexMetaData.numberOfReplicas(), equalTo(3));
assertThat(indexMetaData.settings().getAsMap().size(), equalTo(4));
assertThat(indexMetaData.settings().get("setting1"), equalTo("value1"));
assertThat(indexMetaData.settings().get("setting2"), equalTo("value2"));
assertThat(indexMetaData.mappings().size(), equalTo(0));
indexMetaData = metaData.index("test3");
assertThat(indexMetaData.numberOfShards(), equalTo(1));
assertThat(indexMetaData.numberOfReplicas(), equalTo(2));
assertThat(indexMetaData.settings().getAsMap().size(), equalTo(2));
assertThat(indexMetaData.mappings().size(), equalTo(1));
assertThat(indexMetaData.mappings().get("mapping1"), equalTo(MAPPING_SOURCE1));
indexMetaData = metaData.index("test4");
assertThat(indexMetaData.numberOfShards(), equalTo(1));
assertThat(indexMetaData.numberOfReplicas(), equalTo(2));
assertThat(indexMetaData.settings().getAsMap().size(), equalTo(4));
assertThat(indexMetaData.settings().get("setting1"), equalTo("value1"));
assertThat(indexMetaData.settings().get("setting2"), equalTo("value2"));
assertThat(indexMetaData.mappings().size(), equalTo(2));
assertThat(indexMetaData.mappings().get("mapping1"), equalTo(MAPPING_SOURCE1));
assertThat(indexMetaData.mappings().get("mapping2"), equalTo(MAPPING_SOURCE2));
}
public boolean allocateUnassigned(RoutingAllocation allocation) {
boolean changed = false;
DiscoveryNodes nodes = allocation.nodes();
RoutingNodes routingNodes = allocation.routingNodes();
// First, handle primaries, they must find a place to be allocated on here
Iterator<MutableShardRouting> unassignedIterator = routingNodes.unassigned().iterator();
while (unassignedIterator.hasNext()) {
MutableShardRouting shard = unassignedIterator.next();
if (!shard.primary()) {
continue;
}
// this is an API allocation, ignore since we know there is no data...
if (!routingNodes
.routingTable()
.index(shard.index())
.shard(shard.id())
.primaryAllocatedPostApi()) {
continue;
}
ObjectLongOpenHashMap<DiscoveryNode> nodesState = buildShardStates(nodes, shard);
int numberOfAllocationsFound = 0;
long highestVersion = -1;
Set<DiscoveryNode> nodesWithHighestVersion = Sets.newHashSet();
final boolean[] states = nodesState.allocated;
final Object[] keys = nodesState.keys;
final long[] values = nodesState.values;
for (int i = 0; i < states.length; i++) {
if (!states[i]) {
continue;
}
DiscoveryNode node = (DiscoveryNode) keys[i];
long version = values[i];
// since we don't check in NO allocation, we need to double check here
if (allocation.shouldIgnoreShardForNode(shard.shardId(), node.id())) {
continue;
}
if (version != -1) {
numberOfAllocationsFound++;
if (highestVersion == -1) {
nodesWithHighestVersion.add(node);
highestVersion = version;
} else {
if (version > highestVersion) {
nodesWithHighestVersion.clear();
nodesWithHighestVersion.add(node);
highestVersion = version;
} else if (version == highestVersion) {
nodesWithHighestVersion.add(node);
}
}
}
}
// check if the counts meets the minimum set
int requiredAllocation = 1;
// if we restore from a repository one copy is more then enough
if (shard.restoreSource() == null) {
try {
IndexMetaData indexMetaData = routingNodes.metaData().index(shard.index());
String initialShards =
indexMetaData
.settings()
.get(
INDEX_RECOVERY_INITIAL_SHARDS,
settings.get(INDEX_RECOVERY_INITIAL_SHARDS, this.initialShards));
if ("quorum".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 1) {
requiredAllocation = ((1 + indexMetaData.numberOfReplicas()) / 2) + 1;
}
} else if ("quorum-1".equals(initialShards) || "half".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 2) {
requiredAllocation = ((1 + indexMetaData.numberOfReplicas()) / 2);
}
} else if ("one".equals(initialShards)) {
requiredAllocation = 1;
} else if ("full".equals(initialShards) || "all".equals(initialShards)) {
requiredAllocation = indexMetaData.numberOfReplicas() + 1;
} else if ("full-1".equals(initialShards) || "all-1".equals(initialShards)) {
if (indexMetaData.numberOfReplicas() > 1) {
requiredAllocation = indexMetaData.numberOfReplicas();
}
} else {
requiredAllocation = Integer.parseInt(initialShards);
}
} catch (Exception e) {
logger.warn(
"[{}][{}] failed to derived initial_shards from value {}, ignore allocation for {}",
shard.index(),
shard.id(),
initialShards,
shard);
}
}
// not enough found for this shard, continue...
if (numberOfAllocationsFound < requiredAllocation) {
// if we are restoring this shard we still can allocate
if (shard.restoreSource() == null) {
// we can't really allocate, so ignore it and continue
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: not allocating, number_of_allocated_shards_found [{}], required_number [{}]",
shard.index(),
shard.id(),
numberOfAllocationsFound,
requiredAllocation);
}
} else if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: missing local data, will restore from [{}]",
shard.index(),
shard.id(),
shard.restoreSource());
}
continue;
}
Set<DiscoveryNode> throttledNodes = Sets.newHashSet();
Set<DiscoveryNode> noNodes = Sets.newHashSet();
for (DiscoveryNode discoNode : nodesWithHighestVersion) {
RoutingNode node = routingNodes.node(discoNode.id());
if (node == null) {
continue;
}
Decision decision = allocation.deciders().canAllocate(shard, node, allocation);
if (decision.type() == Decision.Type.THROTTLE) {
throttledNodes.add(discoNode);
} else if (decision.type() == Decision.Type.NO) {
noNodes.add(discoNode);
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: allocating [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
discoNode);
}
// we found a match
changed = true;
// make sure we create one with the version from the recovered state
allocation
.routingNodes()
.assign(new MutableShardRouting(shard, highestVersion), node.nodeId());
unassignedIterator.remove();
// found a node, so no throttling, no "no", and break out of the loop
throttledNodes.clear();
noNodes.clear();
break;
}
}
if (throttledNodes.isEmpty()) {
// if we have a node that we "can't" allocate to, force allocation, since this is our master
// data!
if (!noNodes.isEmpty()) {
DiscoveryNode discoNode = noNodes.iterator().next();
RoutingNode node = routingNodes.node(discoNode.id());
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: forcing allocating [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
discoNode);
}
// we found a match
changed = true;
// make sure we create one with the version from the recovered state
allocation
.routingNodes()
.assign(new MutableShardRouting(shard, highestVersion), node.nodeId());
unassignedIterator.remove();
}
} else {
if (logger.isDebugEnabled()) {
logger.debug(
"[{}][{}]: throttling allocation [{}] to [{}] on primary allocation",
shard.index(),
shard.id(),
shard,
throttledNodes);
}
// we are throttling this, but we have enough to allocate to this node, ignore it for now
unassignedIterator.remove();
routingNodes.ignoredUnassigned().add(shard);
}
}
if (!routingNodes.hasUnassigned()) {
return changed;
}
// Now, handle replicas, try to assign them to nodes that are similar to the one the primary was
// allocated on
unassignedIterator = routingNodes.unassigned().iterator();
while (unassignedIterator.hasNext()) {
MutableShardRouting shard = unassignedIterator.next();
// 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 : 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) {
continue;
}
Map<DiscoveryNode, TransportNodesListShardStoreMetaData.StoreFilesMetaData> shardStores =
buildShardStores(nodes, shard);
long lastSizeMatched = 0;
DiscoveryNode lastDiscoNodeMatched = null;
RoutingNode lastNodeMatched = null;
for (Map.Entry<DiscoveryNode, TransportNodesListShardStoreMetaData.StoreFilesMetaData>
nodeStoreEntry : shardStores.entrySet()) {
DiscoveryNode discoNode = nodeStoreEntry.getKey();
TransportNodesListShardStoreMetaData.StoreFilesMetaData storeFilesMetaData =
nodeStoreEntry.getValue();
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()) {
MutableShardRouting primaryShard = routingNodes.activePrimary(shard);
if (primaryShard != null) {
assert primaryShard.active();
DiscoveryNode primaryNode = nodes.get(primaryShard.currentNodeId());
if (primaryNode != null) {
TransportNodesListShardStoreMetaData.StoreFilesMetaData primaryNodeStore =
shardStores.get(primaryNode);
if (primaryNodeStore != null && primaryNodeStore.allocated()) {
long sizeMatched = 0;
for (StoreFileMetaData storeFileMetaData : storeFilesMetaData) {
if (primaryNodeStore.fileExists(storeFileMetaData.name())
&& primaryNodeStore
.file(storeFileMetaData.name())
.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.remove();
routingNodes.ignoredUnassigned().add(shard);
} 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;
allocation.routingNodes().assign(shard, lastNodeMatched.nodeId());
unassignedIterator.remove();
}
}
}
return changed;
}
private ClusterHealthResponse clusterHealth(ClusterHealthRequest request) {
ClusterState clusterState = clusterService.state();
RoutingTableValidation validation =
clusterState.routingTable().validate(clusterState.metaData());
ClusterHealthResponse response =
new ClusterHealthResponse(clusterName.value(), validation.failures());
response.numberOfNodes = clusterState.nodes().size();
response.numberOfDataNodes = clusterState.nodes().dataNodes().size();
for (String index : clusterState.metaData().concreteIndices(request.indices())) {
IndexRoutingTable indexRoutingTable = clusterState.routingTable().index(index);
IndexMetaData indexMetaData = clusterState.metaData().index(index);
if (indexRoutingTable == null) {
continue;
}
ClusterIndexHealth indexHealth =
new ClusterIndexHealth(
index,
indexMetaData.numberOfShards(),
indexMetaData.numberOfReplicas(),
validation.indexFailures(indexMetaData.index()));
for (IndexShardRoutingTable shardRoutingTable : indexRoutingTable) {
ClusterShardHealth shardHealth = new ClusterShardHealth(shardRoutingTable.shardId().id());
for (ShardRouting shardRouting : shardRoutingTable) {
if (shardRouting.active()) {
shardHealth.activeShards++;
if (shardRouting.relocating()) {
// the shard is relocating, the one he is relocating to will be in initializing state,
// so we don't count it
shardHealth.relocatingShards++;
}
if (shardRouting.primary()) {
shardHealth.primaryActive = true;
}
} else if (shardRouting.initializing()) {
shardHealth.initializingShards++;
} else if (shardRouting.unassigned()) {
shardHealth.unassignedShards++;
}
}
if (shardHealth.primaryActive) {
if (shardHealth.activeShards == shardRoutingTable.size()) {
shardHealth.status = ClusterHealthStatus.GREEN;
} else {
shardHealth.status = ClusterHealthStatus.YELLOW;
}
} else {
shardHealth.status = ClusterHealthStatus.RED;
}
indexHealth.shards.put(shardHealth.id(), shardHealth);
}
for (ClusterShardHealth shardHealth : indexHealth) {
if (shardHealth.primaryActive()) {
indexHealth.activePrimaryShards++;
}
indexHealth.activeShards += shardHealth.activeShards;
indexHealth.relocatingShards += shardHealth.relocatingShards;
indexHealth.initializingShards += shardHealth.initializingShards;
indexHealth.unassignedShards += shardHealth.unassignedShards;
}
// update the index status
indexHealth.status = ClusterHealthStatus.GREEN;
if (!indexHealth.validationFailures().isEmpty()) {
indexHealth.status = ClusterHealthStatus.RED;
} else if (indexHealth
.shards()
.isEmpty()) { // might be since none has been created yet (two phase index creation)
indexHealth.status = ClusterHealthStatus.RED;
} else {
for (ClusterShardHealth shardHealth : indexHealth) {
if (shardHealth.status() == ClusterHealthStatus.RED) {
indexHealth.status = ClusterHealthStatus.RED;
break;
}
if (shardHealth.status() == ClusterHealthStatus.YELLOW) {
indexHealth.status = ClusterHealthStatus.YELLOW;
}
}
}
response.indices.put(indexHealth.index(), indexHealth);
}
for (ClusterIndexHealth indexHealth : response) {
response.activePrimaryShards += indexHealth.activePrimaryShards;
response.activeShards += indexHealth.activeShards;
response.relocatingShards += indexHealth.relocatingShards;
response.initializingShards += indexHealth.initializingShards;
response.unassignedShards += indexHealth.unassignedShards;
}
response.status = ClusterHealthStatus.GREEN;
if (!response.validationFailures().isEmpty()) {
response.status = ClusterHealthStatus.RED;
} else if (clusterState
.blocks()
.hasGlobalBlock(GatewayService.NOT_RECOVERED_FROM_GATEWAY_BLOCK)) {
response.status = ClusterHealthStatus.RED;
} else {
for (ClusterIndexHealth indexHealth : response) {
if (indexHealth.status() == ClusterHealthStatus.RED) {
response.status = ClusterHealthStatus.RED;
break;
}
if (indexHealth.status() == ClusterHealthStatus.YELLOW) {
response.status = ClusterHealthStatus.YELLOW;
}
}
}
return response;
}
