use of org.elasticsearch.cluster.routing.allocation.decider.Decision in project elasticsearch by elastic.
the class ClusterAllocationExplainIT method testAllocationFilteringPreventsShardMove.
public void testAllocationFilteringPreventsShardMove() throws Exception {
logger.info("--> starting 2 nodes");
internalCluster().startNodes(2);
logger.info("--> creating an index with 1 primary and 0 replicas");
createIndexAndIndexData(1, 0);
logger.info("--> setting up allocation filtering to prevent allocation to both nodes");
client().admin().indices().prepareUpdateSettings("idx").setSettings(Settings.builder().put("index.routing.allocation.include._name", "non_existent_node")).get();
boolean includeYesDecisions = randomBoolean();
boolean includeDiskInfo = randomBoolean();
ClusterAllocationExplanation explanation = runExplain(true, includeYesDecisions, includeDiskInfo);
ShardId shardId = explanation.getShard();
boolean isPrimary = explanation.isPrimary();
ShardRoutingState shardRoutingState = explanation.getShardState();
DiscoveryNode currentNode = explanation.getCurrentNode();
UnassignedInfo unassignedInfo = explanation.getUnassignedInfo();
ClusterInfo clusterInfo = explanation.getClusterInfo();
AllocateUnassignedDecision allocateDecision = explanation.getShardAllocationDecision().getAllocateDecision();
MoveDecision moveDecision = explanation.getShardAllocationDecision().getMoveDecision();
// verify shard info
assertEquals("idx", shardId.getIndexName());
assertEquals(0, shardId.getId());
assertTrue(isPrimary);
// verify current node info
assertEquals(ShardRoutingState.STARTED, shardRoutingState);
assertNotNull(currentNode);
// verify unassigned info
assertNull(unassignedInfo);
// verify cluster info
verifyClusterInfo(clusterInfo, includeDiskInfo, 2);
// verify decision object
assertFalse(allocateDecision.isDecisionTaken());
assertTrue(moveDecision.isDecisionTaken());
assertEquals(AllocationDecision.NO, moveDecision.getAllocationDecision());
assertEquals("cannot move shard to another node, even though it is not allowed to remain on its current node", moveDecision.getExplanation());
assertFalse(moveDecision.canRemain());
assertFalse(moveDecision.forceMove());
assertFalse(moveDecision.canRebalanceCluster());
assertNull(moveDecision.getClusterRebalanceDecision());
assertNull(moveDecision.getTargetNode());
assertEquals(0, moveDecision.getCurrentNodeRanking());
// verifying can remain decision object
assertNotNull(moveDecision.getCanRemainDecision());
assertEquals(Decision.Type.NO, moveDecision.getCanRemainDecision().type());
for (Decision d : moveDecision.getCanRemainDecision().getDecisions()) {
if (d.label().equals("filter")) {
assertEquals(Decision.Type.NO, d.type());
assertEquals("node does not match index setting [index.routing.allocation.include] filters [_name:\"non_existent_node\"]", d.getExplanation());
} else {
assertEquals(Decision.Type.YES, d.type());
assertNotNull(d.getExplanation());
}
}
// verify node decisions
assertEquals(1, moveDecision.getNodeDecisions().size());
NodeAllocationResult result = moveDecision.getNodeDecisions().get(0);
assertNotNull(result.getNode());
assertEquals(1, result.getWeightRanking());
assertEquals(AllocationDecision.NO, result.getNodeDecision());
if (includeYesDecisions) {
assertThat(result.getCanAllocateDecision().getDecisions().size(), greaterThan(1));
} else {
assertEquals(1, result.getCanAllocateDecision().getDecisions().size());
}
for (Decision d : result.getCanAllocateDecision().getDecisions()) {
if (d.label().equals("filter")) {
assertEquals(Decision.Type.NO, d.type());
assertEquals("node does not match index setting [index.routing.allocation.include] filters [_name:\"non_existent_node\"]", d.getExplanation());
} else {
assertEquals(Decision.Type.YES, d.type());
assertNotNull(d.getExplanation());
}
}
// verify JSON output
try (XContentParser parser = getParser(explanation)) {
verifyShardInfo(parser, true, includeDiskInfo, ShardRoutingState.STARTED);
parser.nextToken();
assertEquals("can_remain_on_current_node", parser.currentName());
parser.nextToken();
assertEquals(AllocationDecision.NO.toString(), parser.text());
parser.nextToken();
assertEquals("can_remain_decisions", parser.currentName());
verifyDeciders(parser, AllocationDecision.NO);
parser.nextToken();
assertEquals("can_move_to_other_node", parser.currentName());
parser.nextToken();
assertEquals(AllocationDecision.NO.toString(), parser.text());
parser.nextToken();
assertEquals("move_explanation", parser.currentName());
parser.nextToken();
assertEquals("cannot move shard to another node, even though it is not allowed to remain on its current node", parser.text());
verifyNodeDecisions(parser, allNodeDecisions(AllocationDecision.NO, true), includeYesDecisions, false);
assertEquals(Token.END_OBJECT, parser.nextToken());
}
}
use of org.elasticsearch.cluster.routing.allocation.decider.Decision in project elasticsearch by elastic.
the class MoveAllocationCommand method execute.
@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
DiscoveryNode fromDiscoNode = allocation.nodes().resolveNode(fromNode);
DiscoveryNode toDiscoNode = allocation.nodes().resolveNode(toNode);
Decision decision = null;
boolean found = false;
for (ShardRouting shardRouting : allocation.routingNodes().node(fromDiscoNode.getId())) {
if (!shardRouting.shardId().getIndexName().equals(index)) {
continue;
}
if (shardRouting.shardId().id() != shardId) {
continue;
}
found = true;
// TODO we can possibly support also relocating cases, where we cancel relocation and move...
if (!shardRouting.started()) {
if (explain) {
return new RerouteExplanation(this, allocation.decision(Decision.NO, "move_allocation_command", "shard " + shardId + " has not been started"));
}
throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", shard is not started (state = " + shardRouting.state() + "]");
}
RoutingNode toRoutingNode = allocation.routingNodes().node(toDiscoNode.getId());
decision = allocation.deciders().canAllocate(shardRouting, toRoutingNode, allocation);
if (decision.type() == Decision.Type.NO) {
if (explain) {
return new RerouteExplanation(this, decision);
}
throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", from " + fromDiscoNode + ", to " + toDiscoNode + ", since its not allowed, reason: " + decision);
}
if (decision.type() == Decision.Type.THROTTLE) {
// its being throttled, maybe have a flag to take it into account and fail? for now, just do it since the "user" wants it...
}
allocation.routingNodes().relocateShard(shardRouting, toRoutingNode.nodeId(), allocation.clusterInfo().getShardSize(shardRouting, ShardRouting.UNAVAILABLE_EXPECTED_SHARD_SIZE), allocation.changes());
}
if (!found) {
if (explain) {
return new RerouteExplanation(this, allocation.decision(Decision.NO, "move_allocation_command", "shard " + shardId + " not found"));
}
throw new IllegalArgumentException("[move_allocation] can't move " + shardId + ", failed to find it on node " + fromDiscoNode);
}
return new RerouteExplanation(this, decision);
}
use of org.elasticsearch.cluster.routing.allocation.decider.Decision in project elasticsearch by elastic.
the class AllocateReplicaAllocationCommand method execute.
@Override
public RerouteExplanation execute(RoutingAllocation allocation, boolean explain) {
final DiscoveryNode discoNode;
try {
discoNode = allocation.nodes().resolveNode(node);
} catch (IllegalArgumentException e) {
return explainOrThrowRejectedCommand(explain, allocation, e);
}
final RoutingNodes routingNodes = allocation.routingNodes();
RoutingNode routingNode = routingNodes.node(discoNode.getId());
if (routingNode == null) {
return explainOrThrowMissingRoutingNode(allocation, explain, discoNode);
}
final ShardRouting primaryShardRouting;
try {
primaryShardRouting = allocation.routingTable().shardRoutingTable(index, shardId).primaryShard();
} catch (IndexNotFoundException | ShardNotFoundException e) {
return explainOrThrowRejectedCommand(explain, allocation, e);
}
if (primaryShardRouting.unassigned()) {
return explainOrThrowRejectedCommand(explain, allocation, "trying to allocate a replica shard [" + index + "][" + shardId + "], while corresponding primary shard is still unassigned");
}
List<ShardRouting> replicaShardRoutings = allocation.routingTable().shardRoutingTable(index, shardId).replicaShardsWithState(ShardRoutingState.UNASSIGNED);
ShardRouting shardRouting;
if (replicaShardRoutings.isEmpty()) {
return explainOrThrowRejectedCommand(explain, allocation, "all copies of [" + index + "][" + shardId + "] are already assigned. Use the move allocation command instead");
} else {
shardRouting = replicaShardRoutings.get(0);
}
Decision decision = allocation.deciders().canAllocate(shardRouting, routingNode, allocation);
if (decision.type() == Decision.Type.NO) {
// don't use explainOrThrowRejectedCommand to keep the original "NO" decision
if (explain) {
return new RerouteExplanation(this, decision);
}
throw new IllegalArgumentException("[" + name() + "] allocation of [" + index + "][" + shardId + "] on node " + discoNode + " is not allowed, reason: " + decision);
}
initializeUnassignedShard(allocation, routingNodes, routingNode, shardRouting);
return new RerouteExplanation(this, decision);
}
use of org.elasticsearch.cluster.routing.allocation.decider.Decision in project elasticsearch by elastic.
the class ReplicaShardAllocator method canBeAllocatedToAtLeastOneNode.
/**
* Determines if the shard can be allocated on at least one node based on the allocation deciders.
*
* Returns the best allocation decision for allocating the shard on any node (i.e. YES if at least one
* node decided YES, THROTTLE if at least one node decided THROTTLE, and NO if none of the nodes decided
* YES or THROTTLE). If in explain mode, also returns the node-level explanations as the second element
* in the returned tuple.
*/
private Tuple<Decision, Map<String, NodeAllocationResult>> canBeAllocatedToAtLeastOneNode(ShardRouting shard, RoutingAllocation allocation) {
Decision madeDecision = Decision.NO;
final boolean explain = allocation.debugDecision();
Map<String, NodeAllocationResult> nodeDecisions = explain ? new HashMap<>() : null;
for (ObjectCursor<DiscoveryNode> cursor : allocation.nodes().getDataNodes().values()) {
RoutingNode node = allocation.routingNodes().node(cursor.value.getId());
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 && madeDecision.type() != Decision.Type.YES) {
if (explain) {
madeDecision = decision;
} else {
return Tuple.tuple(decision, nodeDecisions);
}
} else if (madeDecision.type() == Decision.Type.NO && decision.type() == Decision.Type.THROTTLE) {
madeDecision = decision;
}
if (explain) {
nodeDecisions.put(node.nodeId(), new NodeAllocationResult(node.node(), null, decision));
}
}
return Tuple.tuple(madeDecision, nodeDecisions);
}
use of org.elasticsearch.cluster.routing.allocation.decider.Decision in project elasticsearch by elastic.
the class FilterAllocationDeciderTests method testFilterInitialRecovery.
public void testFilterInitialRecovery() {
ClusterSettings clusterSettings = new ClusterSettings(Settings.EMPTY, ClusterSettings.BUILT_IN_CLUSTER_SETTINGS);
FilterAllocationDecider filterAllocationDecider = new FilterAllocationDecider(Settings.EMPTY, clusterSettings);
AllocationDeciders allocationDeciders = new AllocationDeciders(Settings.EMPTY, Arrays.asList(filterAllocationDecider, new SameShardAllocationDecider(Settings.EMPTY, clusterSettings), new ReplicaAfterPrimaryActiveAllocationDecider(Settings.EMPTY)));
AllocationService service = new AllocationService(Settings.builder().build(), allocationDeciders, new TestGatewayAllocator(), new BalancedShardsAllocator(Settings.EMPTY), EmptyClusterInfoService.INSTANCE);
ClusterState state = createInitialClusterState(service, Settings.builder().put("index.routing.allocation.initial_recovery._id", "node2").build());
RoutingTable routingTable = state.routingTable();
// we can initally only allocate on node2
assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), INITIALIZING);
assertEquals(routingTable.index("idx").shard(0).shards().get(0).currentNodeId(), "node2");
routingTable = service.applyFailedShard(state, routingTable.index("idx").shard(0).shards().get(0)).routingTable();
state = ClusterState.builder(state).routingTable(routingTable).build();
assertEquals(routingTable.index("idx").shard(0).shards().get(0).state(), UNASSIGNED);
assertNull(routingTable.index("idx").shard(0).shards().get(0).currentNodeId());
// after failing the shard we are unassigned since the node is blacklisted and we can't initialize on the other node
RoutingAllocation allocation = new RoutingAllocation(allocationDeciders, state.getRoutingNodes(), state, null, 0, false);
allocation.debugDecision(true);
Decision.Single decision = (Decision.Single) filterAllocationDecider.canAllocate(routingTable.index("idx").shard(0).primaryShard(), state.getRoutingNodes().node("node2"), allocation);
assertEquals(Type.YES, decision.type());
assertEquals("node passes include/exclude/require filters", decision.getExplanation());
ShardRouting primaryShard = routingTable.index("idx").shard(0).primaryShard();
decision = (Decision.Single) filterAllocationDecider.canAllocate(routingTable.index("idx").shard(0).primaryShard(), state.getRoutingNodes().node("node1"), allocation);
assertEquals(Type.NO, decision.type());
if (primaryShard.recoverySource().getType() == RecoverySource.Type.LOCAL_SHARDS) {
assertEquals("initial allocation of the shrunken index is only allowed on nodes [_id:\"node2\"] that " + "hold a copy of every shard in the index", decision.getExplanation());
} else {
assertEquals("initial allocation of the index is only allowed on nodes [_id:\"node2\"]", decision.getExplanation());
}
state = service.reroute(state, "try allocate again");
routingTable = state.routingTable();
assertEquals(routingTable.index("idx").shard(0).primaryShard().state(), INITIALIZING);
assertEquals(routingTable.index("idx").shard(0).primaryShard().currentNodeId(), "node2");
state = service.applyStartedShards(state, routingTable.index("idx").shard(0).shardsWithState(INITIALIZING));
routingTable = state.routingTable();
// ok now we are started and can be allocated anywhere!! lets see...
// first create another copy
assertEquals(routingTable.index("idx").shard(0).replicaShards().get(0).state(), INITIALIZING);
assertEquals(routingTable.index("idx").shard(0).replicaShards().get(0).currentNodeId(), "node1");
state = service.applyStartedShards(state, routingTable.index("idx").shard(0).replicaShardsWithState(INITIALIZING));
routingTable = state.routingTable();
assertEquals(routingTable.index("idx").shard(0).replicaShards().get(0).state(), STARTED);
assertEquals(routingTable.index("idx").shard(0).replicaShards().get(0).currentNodeId(), "node1");
// now remove the node of the other copy and fail the current
DiscoveryNode node1 = state.nodes().resolveNode("node1");
state = service.deassociateDeadNodes(ClusterState.builder(state).nodes(DiscoveryNodes.builder(state.nodes()).remove("node1")).build(), true, "test");
state = service.applyFailedShard(state, routingTable.index("idx").shard(0).primaryShard());
// now bring back node1 and see it's assigned
state = service.reroute(ClusterState.builder(state).nodes(DiscoveryNodes.builder(state.nodes()).add(node1)).build(), "test");
routingTable = state.routingTable();
assertEquals(routingTable.index("idx").shard(0).primaryShard().state(), INITIALIZING);
assertEquals(routingTable.index("idx").shard(0).primaryShard().currentNodeId(), "node1");
allocation = new RoutingAllocation(allocationDeciders, state.getRoutingNodes(), state, null, 0, false);
allocation.debugDecision(true);
decision = (Decision.Single) filterAllocationDecider.canAllocate(routingTable.index("idx").shard(0).shards().get(0), state.getRoutingNodes().node("node2"), allocation);
assertEquals(Type.YES, decision.type());
assertEquals("node passes include/exclude/require filters", decision.getExplanation());
decision = (Decision.Single) filterAllocationDecider.canAllocate(routingTable.index("idx").shard(0).shards().get(0), state.getRoutingNodes().node("node1"), allocation);
assertEquals(Type.YES, decision.type());
assertEquals("node passes include/exclude/require filters", decision.getExplanation());
}
Aggregations