Github user chtyim commented on a diff in the pull request:
https://github.com/apache/incubator-twill/pull/52#discussion_r34399837
--- Diff:
twill-yarn/src/main/java/org/apache/twill/internal/appmaster/ApplicationMasterService.java
---
@@ -787,4 +804,80 @@ private Resource
createCapability(ResourceSpecification resourceSpec) {
capability.setMemory(resourceSpec.getMemorySize());
return capability;
}
+
+ /**
+ * Attempt to restart some instances from a runnable or some runnables.
+ * @return {@code true} if the message requests restarting some
instances and {@code false} otherwise.
+ */
+ private boolean handleRestartRunnablesInstances(final Message message,
final Runnable completion) {
+ LOG.debug("Check if it should process a restart runnable instances.");
+
+ if (message.getType() != Message.Type.SYSTEM) {
+ return false;
+ }
+
+ Message.Scope messageScope = message.getScope();
+ if (messageScope != Message.Scope.RUNNABLE && messageScope !=
Message.Scope.RUNNABLES) {
+ return false;
+ }
+
+ Command requestCommand = message.getCommand();
+ if
(!Constants.RESTART_ALL_RUNNABLE_INSTANCES.equals(requestCommand.getCommand())
&&
+
!Constants.RESTART_RUNNABLES_INSTANCES.equals(requestCommand.getCommand())) {
+ return false;
+ }
+
+ LOG.debug("Processing restart runnable instances message {}.",
message);
+
+ if (!Strings.isNullOrEmpty(message.getRunnableName()) &&
message.getScope() == Message.Scope.RUNNABLE) {
+ // ... for a runnable ...
+ String runnableName = message.getRunnableName();
+ LOG.debug("Start restarting all runnable {} instances.",
runnableName);
+ restartRunnableInstances(runnableName, null);
+ } else {
+ // ... or maybe some runnables
+ for (Map.Entry<String, String> option :
requestCommand.getOptions().entrySet()) {
+ String runnableName = option.getKey();
+ Set<Integer> restartedInstanceIds =
GSON.fromJson(option.getValue(),
+ new
TypeToken<Set<Integer>>() {}.getType());
+
+ LOG.debug("Start restarting runnable {} instances {}",
runnableName,
+ Arrays.toString(restartedInstanceIds.toArray()));
+ restartRunnableInstances(runnableName, new
LinkedList<>(restartedInstanceIds));
+ }
+ }
+
+ completion.run();
+ return true;
+ }
+
+ /**
+ * Helper method to restart instances of runnables.
+ */
+ private void restartRunnableInstances(String runnableName, @Nullable
Queue<Integer> instanceIds) {
+ LOG.debug("Begin restart runnable {} instances.", runnableName);
+
+ int oldCount = expectedContainers.getExpected(runnableName);
+ LOG.debug("Old instances for runnable {} is {}", runnableName,
oldCount);
+
+ if (instanceIds == null) {
+ LOG.debug("Instance ids is null so stopping all instances for
runnable {}", runnableName);
+
+ // restart all containers for a runnable
+ for (int i = (oldCount - 1); i >= 0; i--) {
--- End diff --
Since you are using the expectedCount, so it's possible that some of the
instances are down (due to whatever reason), meaning it won't really restarting
all containers that are currently running (and some running containers might
not get restarted). Would it be better to take a snapshot of all currently
running instances and restart them (and if failed to restart due to the
container is already down, don't throw)?
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
