use of org.apache.zookeeper.KeeperException.NoNodeException in project helios by spotify.
the class ZooKeeperMasterModel method removeDeploymentGroup.
/**
* Remove a deployment group.
*
* <p>If successful, all ZK nodes associated with the DG will be deleted. Specifically these
* nodes are guaranteed to be non-existent after a successful remove (not all of them might exist
* before, though):
* <ul>
* <li>/config/deployment-groups/[group-name]</li>
* <li>/status/deployment-groups/[group-name]</li>
* <li>/status/deployment-groups/[group-name]/hosts</li>
* <li>/status/deployment-groups/[group-name]/removed</li>
* <li>/status/deployment-group-tasks/[group-name]</li>
* </ul>
* If the operation fails no ZK nodes will be removed.
*
* @throws DeploymentGroupDoesNotExistException If the DG does not exist.
*/
@Override
public void removeDeploymentGroup(final String name) throws DeploymentGroupDoesNotExistException {
log.info("removing deployment-group: name={}", name);
final ZooKeeperClient client = provider.get("removeDeploymentGroup");
try {
client.ensurePath(Paths.configDeploymentGroups());
client.ensurePath(Paths.statusDeploymentGroups());
client.ensurePath(Paths.statusDeploymentGroupTasks());
final List<ZooKeeperOperation> operations = Lists.newArrayList();
final List<String> paths = ImmutableList.of(Paths.configDeploymentGroup(name), Paths.statusDeploymentGroup(name), Paths.statusDeploymentGroupHosts(name), Paths.statusDeploymentGroupRemovedHosts(name), Paths.statusDeploymentGroupTasks(name));
// DGs to become slower and spam logs with errors so we want to avoid it.
for (final String path : paths) {
if (client.exists(path) == null) {
operations.add(create(path));
}
}
for (final String path : Lists.reverse(paths)) {
operations.add(delete(path));
}
client.transaction(operations);
} catch (final NoNodeException e) {
throw new DeploymentGroupDoesNotExistException(name);
} catch (final KeeperException e) {
throw new HeliosRuntimeException("removing deployment-group " + name + " failed", e);
}
}
use of org.apache.zookeeper.KeeperException.NoNodeException in project helios by spotify.
the class ZooKeeperMasterModel method addJob.
/**
* Adds a job into the configuration.
*/
@Override
public void addJob(final Job job) throws JobExistsException {
log.info("adding job: {}", job);
final JobId id = job.getId();
final UUID operationId = UUID.randomUUID();
final String creationPath = Paths.configJobCreation(id, operationId);
final ZooKeeperClient client = provider.get("addJob");
try {
try {
client.ensurePath(Paths.historyJob(id));
client.transaction(create(Paths.configJob(id), job), create(Paths.configJobRefShort(id), id), create(Paths.configJobHosts(id)), create(creationPath), // change down the tree. Effectively, make it that version == cVersion.
set(Paths.configJobs(), UUID.randomUUID().toString().getBytes()));
} catch (final NodeExistsException e) {
if (client.exists(creationPath) != null) {
// The job was created, we're done here
return;
}
throw new JobExistsException(id.toString());
}
} catch (NoNodeException e) {
throw new HeliosRuntimeException("adding job " + job + " failed due to missing ZK path: " + e.getPath(), e);
} catch (final KeeperException e) {
throw new HeliosRuntimeException("adding job " + job + " failed", e);
}
}
use of org.apache.zookeeper.KeeperException.NoNodeException in project helios by spotify.
the class ZooKeeperMasterModel method stopDeploymentGroup.
@Override
public void stopDeploymentGroup(final String deploymentGroupName) throws DeploymentGroupDoesNotExistException {
checkNotNull(deploymentGroupName, "name");
log.info("stop deployment-group: name={}", deploymentGroupName);
final ZooKeeperClient client = provider.get("stopDeploymentGroup");
// Delete deployment group tasks (if any) and set DG state to FAILED
final DeploymentGroupStatus status = DeploymentGroupStatus.newBuilder().setState(FAILED).setError("Stopped by user").build();
final String statusPath = Paths.statusDeploymentGroup(deploymentGroupName);
final String tasksPath = Paths.statusDeploymentGroupTasks(deploymentGroupName);
try {
client.ensurePath(Paths.statusDeploymentGroupTasks());
final List<ZooKeeperOperation> operations = Lists.newArrayList();
// NOTE: This remove operation is racey. If tasks exist and the rollout finishes before the
// delete() is executed then this will fail. Conversely, if it doesn't exist but is created
// before the transaction is executed it will also fail. This is annoying for users, but at
// least means we won't have inconsistent state.
//
// That the set() is first in the list of operations is important because of the
// kludgy error checking we do below to disambiguate "doesn't exist" failures from the race
// condition mentioned below.
operations.add(set(statusPath, status));
final Stat tasksStat = client.exists(tasksPath);
if (tasksStat != null) {
operations.add(delete(tasksPath));
} else {
// There doesn't seem to be a "check that node doesn't exist" operation so we
// do a create and a delete on the same path to emulate it.
operations.add(create(tasksPath));
operations.add(delete(tasksPath));
}
client.transaction(operations);
} catch (final NoNodeException e) {
// Yes, the way you figure out which operation in a transaction failed is retarded.
if (((OpResult.ErrorResult) e.getResults().get(0)).getErr() == KeeperException.Code.NONODE.intValue()) {
throw new DeploymentGroupDoesNotExistException(deploymentGroupName);
} else {
throw new HeliosRuntimeException("stop deployment-group " + deploymentGroupName + " failed due to a race condition, please retry", e);
}
} catch (final KeeperException e) {
throw new HeliosRuntimeException("stop deployment-group " + deploymentGroupName + " failed", e);
}
}
use of org.apache.zookeeper.KeeperException.NoNodeException in project helios by spotify.
the class JobHistoryReaper method processItem.
@Override
void processItem(final String jobId) {
log.info("Deciding whether to reap job history for job {}", jobId);
final JobId id = JobId.fromString(jobId);
final Job job = masterModel.getJob(id);
if (job == null) {
try {
client.deleteRecursive(Paths.historyJob(id));
log.info("Reaped job history for job {}", jobId);
} catch (NoNodeException ignored) {
// Something deleted the history right before we got to it. Ignore and keep going.
} catch (KeeperException e) {
log.warn("error reaping job history for job {}", jobId, e);
}
}
}
use of org.apache.zookeeper.KeeperException.NoNodeException in project helios by spotify.
the class ZooKeeperMasterModel method getJobs.
/**
* Returns a {@link Map} of {@link JobId} to {@link Job} objects for all of the jobs known.
*/
@Override
public Map<JobId, Job> getJobs() {
log.debug("getting jobs");
final String folder = Paths.configJobs();
final ZooKeeperClient client = provider.get("getJobs");
try {
final List<String> ids;
try {
ids = client.getChildren(folder);
} catch (NoNodeException e) {
return Maps.newHashMap();
}
final Map<JobId, Job> descriptors = Maps.newHashMap();
for (final String id : ids) {
final JobId jobId = JobId.fromString(id);
final String path = Paths.configJob(jobId);
try {
final byte[] data = client.getData(path);
final Job descriptor = parse(data, Job.class);
descriptors.put(descriptor.getId(), descriptor);
} catch (NoNodeException e) {
// Ignore, the job was deleted before we had a chance to read it.
log.debug("Ignoring deleted job {}", jobId);
}
}
return descriptors;
} catch (KeeperException | IOException e) {
throw new HeliosRuntimeException("getting jobs failed", e);
}
}
Aggregations