@Override
public String getRpcAddress() {
String sensorName = getConfig(RPC_ADDRESS_SENSOR);
if (Strings.isNonBlank(sensorName))
return Entities.submit(
this,
DependentConfiguration.attributeWhenReady(this, Sensors.newStringSensor(sensorName)))
.getUnchecked();
return "0.0.0.0";
}
@Override
public String getBroadcastAddress() {
String sensorName = getConfig(BROADCAST_ADDRESS_SENSOR);
if (Strings.isNonBlank(sensorName))
return Entities.submit(
this,
DependentConfiguration.attributeWhenReady(this, Sensors.newStringSensor(sensorName)))
.getUnchecked();
String snitchName = getConfig(CassandraNode.ENDPOINT_SNITCH_NAME);
if (snitchName.equals("Ec2MultiRegionSnitch") || snitchName.contains("MultiCloudSnitch")) {
// http://www.datastax.com/documentation/cassandra/2.0/mobile/cassandra/architecture/architectureSnitchEC2MultiRegion_c.html
// describes that the listen_address is set to the private IP, and the broadcast_address is
// set to the public IP.
return getAttribute(CassandraNode.ADDRESS);
} else if (!getDriver().isClustered()) {
return getListenAddress();
} else {
// In other situations, prefer the hostname, so other regions can see it
// *Unless* hostname resolves at the target to a local-only interface which is different to
// ADDRESS
// (workaround for issue deploying to localhost)
String hostname = getAttribute(CassandraNode.HOSTNAME);
try {
String resolvedAddress = getDriver().getResolvedAddress(hostname);
if (resolvedAddress == null) {
log.debug(
"Cassandra using broadcast address "
+ getListenAddress()
+ " for "
+ this
+ " because hostname "
+ hostname
+ " could not be resolved at remote machine");
return getListenAddress();
}
if (resolvedAddress.equals("127.0.0.1")) {
log.debug(
"Cassandra using broadcast address "
+ getListenAddress()
+ " for "
+ this
+ " because hostname "
+ hostname
+ " resolves to 127.0.0.1");
return getListenAddress();
}
return hostname;
} catch (Exception e) {
Exceptions.propagateIfFatal(e);
log.warn("Error resolving hostname " + hostname + " for " + this + ": " + e, e);
return hostname;
}
}
}
@Override
public String getListenAddress() {
String sensorName = getConfig(LISTEN_ADDRESS_SENSOR);
if (Strings.isNonBlank(sensorName))
return Entities.submit(
this,
DependentConfiguration.attributeWhenReady(this, Sensors.newStringSensor(sensorName)))
.getUnchecked();
String subnetAddress = getAttribute(CassandraNode.SUBNET_ADDRESS);
return Strings.isNonBlank(subnetAddress) ? subnetAddress : getAttribute(CassandraNode.ADDRESS);
}
protected void doStop(ConfigBag parameters, Callable<StopMachineDetails<Integer>> stopTask) {
preStopConfirmCustom();
log.info("Stopping {} in {}", entity(), entity().getLocations());
StopMode stopMachineMode = getStopMachineMode(parameters);
StopMode stopProcessMode = parameters.get(StopSoftwareParameters.STOP_PROCESS_MODE);
DynamicTasks.queue("pre-stop", new PreStopCustomTask());
// BROOKLYN-263:
// With this change the stop effector will wait for Location to provision so it can terminate
// the machine, if a provisioning request is in-progress.
//
// The ProvisionMachineTask stores transient internal state in PROVISIONING_TASK_STATE and
// PROVISIONED_MACHINE: it records when the provisioning is running and when done; and it
// records the final machine. We record the machine in the internal sensor (rather than
// just relying on getLocations) because the latter is set much later in the start()
// process.
//
// This code is a big improvement (previously there was a several-minute window in some
// clouds where a call to stop() would leave the machine running).
//
// However, there are still races. If the start() code has not yet reached the call to
// location.obtain() then we won't wait, and the start() call won't know to abort. It's
// fiddly to get that right, because we need to cope with restart() - so we mustn't leave
// any state behind that will interfere with subsequent sequential calls to start().
// There is some attempt to handle it by ProvisionMachineTask checking if the expectedState
// is stopping/stopped.
Maybe<MachineLocation> machine = Machines.findUniqueMachineLocation(entity().getLocations());
ProvisioningTaskState provisioningState =
entity().sensors().get(AttributesInternal.INTERNAL_PROVISIONING_TASK_STATE);
if (machine.isAbsent() && provisioningState == ProvisioningTaskState.RUNNING) {
Duration maxWait = entity().config().get(STOP_WAIT_PROVISIONING_TIMEOUT);
log.info(
"When stopping {}, waiting for up to {} for the machine to finish provisioning, before terminating it",
entity(),
maxWait);
boolean success =
Repeater.create("Wait for a machine to appear")
.until(
new Callable<Boolean>() {
@Override
public Boolean call() throws Exception {
ProvisioningTaskState state =
entity()
.sensors()
.get(AttributesInternal.INTERNAL_PROVISIONING_TASK_STATE);
return (state != ProvisioningTaskState.RUNNING);
}
})
.backoffTo(Duration.FIVE_SECONDS)
.limitTimeTo(maxWait)
.run();
if (!success) {
log.warn(
"When stopping {}, timed out after {} waiting for the machine to finish provisioning - machine may we left running",
entity(),
maxWait);
}
machine = Maybe.ofDisallowingNull(entity().sensors().get(INTERNAL_PROVISIONED_MACHINE));
}
entity().sensors().remove(AttributesInternal.INTERNAL_PROVISIONING_TASK_STATE);
entity().sensors().remove(INTERNAL_PROVISIONED_MACHINE);
Task<List<?>> stoppingProcess = null;
if (canStop(stopProcessMode, entity())) {
stoppingProcess =
Tasks.parallel(
"stopping",
Tasks.create("stopping (process)", new StopProcessesAtMachineTask()),
Tasks.create("stopping (feeds)", new StopFeedsAtMachineTask()));
DynamicTasks.queue(stoppingProcess);
}
Task<StopMachineDetails<Integer>> stoppingMachine = null;
if (canStop(stopMachineMode, machine.isAbsent())) {
// Release this machine (even if error trying to stop process - we rethrow that after)
Map<String, Object> stopMachineFlags = MutableMap.of();
if (Entitlements.getEntitlementContext() != null) {
stopMachineFlags.put(
"tags",
MutableSet.of(
BrooklynTaskTags.tagForEntitlement(Entitlements.getEntitlementContext())));
}
Task<StopMachineDetails<Integer>> stopMachineTask =
Tasks.<StopMachineDetails<Integer>>builder()
.displayName("stopping (machine)")
.body(stopTask)
.flags(stopMachineFlags)
.build();
stoppingMachine = DynamicTasks.queue(stopMachineTask);
DynamicTasks.drain(entity().getConfig(STOP_PROCESS_TIMEOUT), false);
// shutdown the machine if stopping process fails or takes too long
synchronized (stoppingMachine) {
// task also used as mutex by DST when it submits it; ensure it only submits once!
if (!stoppingMachine.isSubmitted()) {
// force the stoppingMachine task to run by submitting it here
StringBuilder msg =
new StringBuilder("Submitting machine stop early in background for ")
.append(entity());
if (stoppingProcess == null) {
msg.append(". Process stop skipped, pre-stop not finished?");
} else {
msg.append(" because process stop has ")
.append((stoppingProcess.isDone() ? "finished abnormally" : "not finished"));
}
log.warn(msg.toString());
Entities.submit(entity(), stoppingMachine);
}
}
}
try {
// This maintains previous behaviour of silently squashing any errors on the stoppingProcess
// task if the
// stoppingMachine exits with a nonzero value
boolean checkStopProcesses =
(stoppingProcess != null
&& (stoppingMachine == null || stoppingMachine.get().value == 0));
if (checkStopProcesses) {
// TODO we should test for destruction above, not merely successful "stop", as things like
// localhost and ssh won't be destroyed
DynamicTasks.waitForLast();
if (machine.isPresent()) {
// throw early errors *only if* there is a machine and we have not destroyed it
stoppingProcess.get();
}
}
} catch (Throwable e) {
ServiceStateLogic.setExpectedState(entity(), Lifecycle.ON_FIRE);
Exceptions.propagate(e);
}
entity().sensors().set(SoftwareProcess.SERVICE_UP, false);
ServiceStateLogic.setExpectedState(entity(), Lifecycle.STOPPED);
DynamicTasks.queue("post-stop", new PostStopCustomTask());
if (log.isDebugEnabled()) log.debug("Stopped software process entity " + entity());
}