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()).allocatedPostApi()) {
continue;
}
TObjectLongHashMap<DiscoveryNode> nodesState = buildShardStates(nodes, shard);
int numberOfAllocationsFound = 0;
long highestVersion = -1;
Set<DiscoveryNode> nodesWithHighestVersion = Sets.newHashSet();
for (TObjectLongIterator<DiscoveryNode> it = nodesState.iterator(); it.hasNext(); ) {
it.advance();
DiscoveryNode node = it.key();
long version = it.value();
// 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;
try {
IndexMetaData indexMetaData = routingNodes.metaData().index(shard.index());
String initialShards = indexMetaData.settings().get("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 ("full".equals(initialShards)) {
requiredAllocation = indexMetaData.numberOfReplicas() + 1;
} else if ("full-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) {
// 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);
}
continue;
}
Set<DiscoveryNode> throttledNodes = Sets.newHashSet();
Set<DiscoveryNode> noNodes = Sets.newHashSet();
for (DiscoveryNode discoNode : nodesWithHighestVersion) {
RoutingNode node = routingNodes.node(discoNode.id());
Decision decision = nodeAllocations.canAllocate(shard, node, allocation);
if (decision == NodeAllocation.Decision.THROTTLE) {
throttledNodes.add(discoNode);
} else if (decision == Decision.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
node.add(new MutableShardRouting(shard, highestVersion));
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
node.add(new MutableShardRouting(shard, highestVersion));
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 (DiscoveryNode discoNode : nodes.dataNodes().values()) {
RoutingNode node = routingNodes.node(discoNode.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
if (nodeAllocations.canAllocate(shard, node, allocation).allocate()) {
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
if (nodeAllocations.canAllocate(shard, node, allocation) == Decision.NO) {
continue;
}
// if it is already allocated, we can't assign to it...
if (storeFilesMetaData.allocated()) {
continue;
}
if (!shard.primary()) {
MutableShardRouting primaryShard = routingNodes.findPrimaryForReplica(shard);
if (primaryShard != null && 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;