@Override
public void run() {
logger.trace("CleanerRunnable starting: hashCode={}", this.hashCode());
PresenceService presenceService = getContext().getService("presence");
CoordinationService coordinationService = getContext().getService("coord");
ZkClient zkClient = getContext().getZkClient();
PathScheme pathScheme = getContext().getPathScheme();
// list all services in cluster
List<String> clusterIds = presenceService.getClusters();
for (String clusterId : clusterIds) {
// only proceed if in cluster
if (!presenceService.isMemberOf(clusterId)
|| clusterId.equals(getContext().getPathScheme().getFrameworkClusterId())) {
continue;
}
List<String> serviceIds = presenceService.getServices(clusterId);
for (String serviceId : serviceIds) {
logger.trace(
"Checking data nodes expiry: clusterId={}; serviceId={}", clusterId, serviceId);
// only proceed if in service
if (!presenceService.isMemberOf(clusterId, serviceId)) {
continue;
}
long currentTimestamp = System.currentTimeMillis();
// get lock for a service
DistributedLock lock =
coordinationService.getLock("reign", "metrics-" + clusterId + "-" + serviceId);
if (!lock.tryLock()) {
continue;
}
String dataPath = null;
try {
// get all data nodes for a service
String dataParentPath =
pathScheme.getAbsolutePath(
PathType.METRICS, pathScheme.joinTokens(clusterId, serviceId));
List<String> dataNodes = zkClient.getChildren(dataParentPath, false);
// remove all nodes that are older than rotation
// interval
for (String dataNode : dataNodes) {
try {
logger.trace(
"Checking data node expiry: clusterId={}; serviceId={}; nodeId={}",
clusterId,
serviceId,
dataNode);
dataPath =
pathScheme.getAbsolutePath(
PathType.METRICS, pathScheme.joinTokens(clusterId, serviceId, dataNode));
MetricsData metricsData = getMetricsFromDataNode(clusterId, serviceId, dataNode);
if (metricsData == null) {
logger.warn(
"Removing unrecognized/corrupted/deprecated data node: path={}", dataPath);
zkClient.delete(dataPath, -1);
continue;
}
// keep last few hours worth of data
long millisToExpiry =
millisToExpiry(metricsData, currentTimestamp - (86400000 / 6));
// delete data that is older than some threshold
boolean dataTooOld =
currentTimestamp - metricsData.getIntervalStartTimestamp() > 86400000;
// delete old and expired data
if (millisToExpiry <= 0 || dataTooOld) {
logger.info(
"Removing expired data node: path={}; millisToExpiry={}",
dataPath,
millisToExpiry);
zkClient.delete(dataPath, -1);
} else {
logger.trace(
"Data node is not yet expired: path={}; millisToExpiry={}",
dataPath,
millisToExpiry);
}
} catch (Exception e) {
logger.warn(
"Error trying to clean up data directory for service: clusterId="
+ clusterId
+ "; serviceId="
+ serviceId
+ "; dataPath="
+ dataPath
+ ": "
+ e,
e);
} // try
} // for
} catch (KeeperException e) {
if (e.code() != KeeperException.Code.NONODE) {
logger.warn(
"Error trying to clean up data directory for service: clusterId="
+ clusterId
+ "; serviceId="
+ serviceId
+ "; dataPath="
+ dataPath
+ ": "
+ e,
e);
}
} catch (Exception e) {
logger.warn(
"Error trying to clean up data directory for service: clusterId="
+ clusterId
+ "; serviceId="
+ serviceId
+ "; dataPath="
+ dataPath
+ ": "
+ e,
e);
} finally {
lock.unlock();
lock.destroy();
} // try
} // for service
} // for cluster
} // run()
