zentol commented on a change in pull request #10089: [FLINK-12342][yarn] Remove
container requests in order to reduce excess containers
URL: https://github.com/apache/flink/pull/10089#discussion_r343163521
##########
File path:
flink-yarn/src/main/java/org/apache/flink/yarn/YarnResourceManager.java
##########
@@ -370,45 +370,24 @@ public void onContainersCompleted(final
List<ContainerStatus> statuses) {
@Override
public void onContainersAllocated(List<Container> containers) {
runAsync(() -> {
+ log.info("Received {} containers with {} pending
container requests.", containers.size(), numPendingContainerRequests);
final Collection<AMRMClient.ContainerRequest>
pendingRequests = getPendingRequests();
final Iterator<AMRMClient.ContainerRequest>
pendingRequestsIterator = pendingRequests.iterator();
- for (Container container : containers) {
- log.info(
- "Received new container: {} - Remaining
pending container requests: {}",
- container.getId(),
- numPendingContainerRequests);
+ final int numAcceptedContainers =
Math.min(containers.size(), numPendingContainerRequests);
+ final List<Container> requiredContainers =
containers.subList(0, numAcceptedContainers);
+ final List<Container> excessContainers =
containers.subList(numAcceptedContainers, containers.size());
- if (numPendingContainerRequests > 0) {
-
removeContainerRequest(pendingRequestsIterator.next());
-
- final String containerIdStr =
container.getId().toString();
- final ResourceID resourceId = new
ResourceID(containerIdStr);
-
- workerNodeMap.put(resourceId, new
YarnWorkerNode(container));
-
- try {
- // Context information used to
start a TaskExecutor Java process
- ContainerLaunchContext
taskExecutorLaunchContext = createTaskExecutorLaunchContext(
- container.getResource(),
- containerIdStr,
-
container.getNodeId().getHost());
+ for (int i = 0; i < requiredContainers.size(); i++) {
+
removeContainerRequest(pendingRequestsIterator.next());
+ }
-
nodeManagerClient.startContainer(container, taskExecutorLaunchContext);
- } catch (Throwable t) {
- log.error("Could not start
TaskManager in container {}.", container.getId(), t);
+ for (Container excessContainer : excessContainers) {
+ returnExcessContainer(excessContainer);
+ }
- // release the failed container
-
workerNodeMap.remove(resourceId);
-
resourceManagerClient.releaseAssignedContainer(container.getId());
- // and ask for a new one
-
requestYarnContainerIfRequired();
- }
- } else {
- // return the excessive containers
- log.info("Returning excess container
{}.", container.getId());
-
resourceManagerClient.releaseAssignedContainer(container.getId());
- }
+ for (Container requiredContainer : requiredContainers) {
+ startTaskExecutorInContainer(requiredContainer);
Review comment:
Do we have any info on how often this call fails? (we could reuse an excess
container if this fails, which would save us from having to re-allocate another
one)
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
