private synchronized void getZkRunning() throws Exception {
LOG.debug("Reading " + parentZnode + Constants.DEFAULT_ZOOKEEPER_ZNODE_SERVERS_RUNNING);
List<String> children =
getChildren(
parentZnode + Constants.DEFAULT_ZOOKEEPER_ZNODE_SERVERS_RUNNING, new RunningWatcher());
if (!children.isEmpty()) {
for (String child : children) {
// If stop-wms.sh is executed and WMS_MANAGES_ZK then zookeeper
// is stopped abruptly.
// Second scenario is when ZooKeeper fails for some reason
// regardless of whether WMS
// manages it. When either happens the WmsServer running znodes
// still exist in ZooKeeper
// and we see them at next startup. When they eventually timeout
// we get node deleted events for a server that no longer
// exists. So, only recognize
// WmsServer running znodes that have timestamps after last
// WmsMaster startup.
Scanner scn = new Scanner(child);
scn.useDelimiter(":");
String hostName = scn.next();
String instance = scn.next();
int infoPort = Integer.parseInt(scn.next());
long serverStartTimestamp = Long.parseLong(scn.next());
scn.close();
if (serverStartTimestamp < startupTimestamp) continue;
if (!runningServers.contains(child)) {
LOG.debug("Watching running [" + child + "]");
zkc.exists(
parentZnode + Constants.DEFAULT_ZOOKEEPER_ZNODE_SERVERS_RUNNING + "/" + child,
new RunningWatcher());
runningServers.add(child);
}
}
metrics.setTotalRunning(runningServers.size());
} else {
metrics.setTotalRunning(0);
}
}
private synchronized void restartServer(String znodePath) throws Exception {
String child =
znodePath.replace(
parentZnode + Constants.DEFAULT_ZOOKEEPER_ZNODE_SERVERS_RUNNING + "/", "");
Scanner scn = new Scanner(child);
scn.useDelimiter(":");
String hostName = scn.next();
String instance = scn.next();
int infoPort = Integer.parseInt(scn.next());
long serverStartTimestamp = Long.parseLong(scn.next());
scn.close();
LOG.error("WmsServer [" + hostName + ":" + instance + "] failed.");
if (runningServers.contains(child)) {
LOG.debug("Found [" + child + "], deleting from running servers list");
runningServers.remove(child);
metrics.setTotalRunning(runningServers.size());
}
RestartHandler handler = new RestartHandler(child);
restartQueue.add(handler);
}