/**
* Deletes a shard data directory. Note: this method assumes that the shard lock is acquired. This
* method will also attempt to acquire the write locks for the shard's paths before deleting the
* data, but this is best effort, as the lock is released before the deletion happens in order to
* allow the folder to be deleted
*
* @param lock the shards lock
* @throws IOException if an IOException occurs
* @throws ElasticsearchException if the write.lock is not acquirable
*/
public void deleteShardDirectoryUnderLock(ShardLock lock, IndexSettings indexSettings)
throws IOException {
final ShardId shardId = lock.getShardId();
assert isShardLocked(shardId) : "shard " + shardId + " is not locked";
final Path[] paths = availableShardPaths(shardId);
logger.trace("acquiring locks for {}, paths: [{}]", shardId, paths);
acquireFSLockForPaths(indexSettings, paths);
IOUtils.rm(paths);
if (indexSettings.hasCustomDataPath()) {
Path customLocation = resolveCustomLocation(indexSettings, shardId);
logger.trace("acquiring lock for {}, custom path: [{}]", shardId, customLocation);
acquireFSLockForPaths(indexSettings, customLocation);
logger.trace("deleting custom shard {} directory [{}]", shardId, customLocation);
IOUtils.rm(customLocation);
}
logger.trace("deleted shard {} directory, paths: [{}]", shardId, paths);
assert FileSystemUtils.exists(paths) == false;
}
private void rerouteWithAllocateLocalGateway(Settings commonSettings) throws Exception {
logger.info("--> starting 2 nodes");
String node_1 = cluster().startNode(commonSettings);
cluster().startNode(commonSettings);
assertThat(cluster().size(), equalTo(2));
ClusterHealthResponse healthResponse =
client().admin().cluster().prepareHealth().setWaitForNodes("2").execute().actionGet();
assertThat(healthResponse.isTimedOut(), equalTo(false));
logger.info("--> create an index with 1 shard, 1 replica, nothing should allocate");
client()
.admin()
.indices()
.prepareCreate("test")
.setSettings(settingsBuilder().put("index.number_of_shards", 1))
.execute()
.actionGet();
ClusterState state = client().admin().cluster().prepareState().execute().actionGet().getState();
assertThat(state.routingNodes().unassigned().size(), equalTo(2));
logger.info("--> explicitly allocate shard 1, actually allocating, no dry run");
state =
client()
.admin()
.cluster()
.prepareReroute()
.setExplain(randomBoolean())
.add(new AllocateAllocationCommand(new ShardId("test", 0), node_1, true))
.execute()
.actionGet()
.getState();
assertThat(state.routingNodes().unassigned().size(), equalTo(1));
assertThat(
state.routingNodes().node(state.nodes().resolveNode(node_1).id()).get(0).state(),
equalTo(ShardRoutingState.INITIALIZING));
healthResponse =
client()
.admin()
.cluster()
.prepareHealth()
.setWaitForEvents(Priority.LANGUID)
.setWaitForYellowStatus()
.execute()
.actionGet();
assertThat(healthResponse.isTimedOut(), equalTo(false));
logger.info("--> get the state, verify shard 1 primary allocated");
state = client().admin().cluster().prepareState().execute().actionGet().getState();
assertThat(state.routingNodes().unassigned().size(), equalTo(1));
assertThat(
state.routingNodes().node(state.nodes().resolveNode(node_1).id()).get(0).state(),
equalTo(ShardRoutingState.STARTED));
client()
.prepareIndex("test", "type", "1")
.setSource("field", "value")
.setRefresh(true)
.execute()
.actionGet();
logger.info("--> closing all nodes");
File[] shardLocation =
cluster().getInstance(NodeEnvironment.class, node_1).shardLocations(new ShardId("test", 0));
assertThat(
FileSystemUtils.exists(shardLocation), equalTo(true)); // make sure the data is there!
cluster()
.closeNonSharedNodes(false); // don't wipe data directories the index needs to be there!
logger.info("--> deleting the shard data [{}] ", Arrays.toString(shardLocation));
assertThat(
FileSystemUtils.exists(shardLocation),
equalTo(true)); // verify again after cluster was shut down
assertThat(FileSystemUtils.deleteRecursively(shardLocation), equalTo(true));
logger.info(
"--> starting nodes back, will not allocate the shard since it has no data, but the index will be there");
node_1 = cluster().startNode(commonSettings);
cluster().startNode(commonSettings);
// wait a bit for the cluster to realize that the shard is not there...
// TODO can we get around this? the cluster is RED, so what do we wait for?
client().admin().cluster().prepareReroute().get();
assertThat(
client()
.admin()
.cluster()
.prepareHealth()
.setWaitForNodes("2")
.execute()
.actionGet()
.getStatus(),
equalTo(ClusterHealthStatus.RED));
logger.info("--> explicitly allocate primary");
state =
client()
.admin()
.cluster()
.prepareReroute()
.setExplain(randomBoolean())
.add(new AllocateAllocationCommand(new ShardId("test", 0), node_1, true))
.execute()
.actionGet()
.getState();
assertThat(state.routingNodes().unassigned().size(), equalTo(1));
assertThat(
state.routingNodes().node(state.nodes().resolveNode(node_1).id()).get(0).state(),
equalTo(ShardRoutingState.INITIALIZING));
healthResponse =
client()
.admin()
.cluster()
.prepareHealth()
.setWaitForEvents(Priority.LANGUID)
.setWaitForYellowStatus()
.execute()
.actionGet();
assertThat(healthResponse.isTimedOut(), equalTo(false));
logger.info("--> get the state, verify shard 1 primary allocated");
state = client().admin().cluster().prepareState().execute().actionGet().getState();
assertThat(state.routingNodes().unassigned().size(), equalTo(1));
assertThat(
state.routingNodes().node(state.nodes().resolveNode(node_1).id()).get(0).state(),
equalTo(ShardRoutingState.STARTED));
}