Github user tillrohrmann commented on a diff in the pull request:
https://github.com/apache/flink/pull/2389#discussion_r75682445
--- Diff:
flink-runtime/src/main/java/org/apache/flink/runtime/rpc/resourcemanager/ResourceManager.java
---
@@ -94,26 +130,218 @@ public RegistrationResponse apply(final
JobMasterGateway jobMasterGateway) {
* @return Slot assignment
*/
@RpcMethod
- public SlotAssignment requestSlot(SlotRequest slotRequest) {
- System.out.println("SlotRequest: " + slotRequest);
- return new SlotAssignment();
+ public AcknowledgeSlotRequest requestSlot(SlotRequest slotRequest) {
+ // TODO slotManager.requestSlot(slotRequest);
+ return new
AcknowledgeSlotRequest(slotRequest.getAllocationID());
}
/**
- *
- * @param resourceManagerLeaderId The fencing token for the
ResourceManager leader
- * @param taskExecutorAddress The address of the TaskExecutor that
registers
- * @param resourceID The resource ID of the TaskExecutor
that registers
+ * @param resourceManagerLeaderId The fencing token for the
ResourceManager leader
+ * @param taskExecutorAddress The address of the TaskExecutor that
registers
+ * @param resourceID The resource ID of the TaskExecutor
that registers
*
* @return The response by the ResourceManager.
*/
@RpcMethod
- public org.apache.flink.runtime.rpc.registration.RegistrationResponse
registerTaskExecutor(
- UUID resourceManagerLeaderId,
- String taskExecutorAddress,
- ResourceID resourceID) {
+ public
Future<org.apache.flink.runtime.rpc.registration.RegistrationResponse>
registerTaskExecutor(
+ final UUID resourceManagerLeaderId,
+ final String taskExecutorAddress,
+ final ResourceID resourceID
+ ) {
+ log.info("received register from taskExecutor {}, address {}",
resourceID, taskExecutorAddress);
+ Future<TaskExecutorGateway> taskExecutorFuture =
+ getRpcService().connect(taskExecutorAddress,
TaskExecutorGateway.class);
+
+ return taskExecutorFuture.map(new Mapper<TaskExecutorGateway,
org.apache.flink.runtime.rpc.registration.RegistrationResponse>() {
+ @Override
+ public
org.apache.flink.runtime.rpc.registration.RegistrationResponse apply(final
TaskExecutorGateway taskExecutorGateway) {
+ // decline registration if resourceManager
cannot connect to the taskExecutor using the given address
+ if(taskExecutorGateway == null) {
+ log.warn("resourceManager {} decline
registration from the taskExecutor {}, cannot connect to it using given address
{} ",
+ getAddress(), resourceID,
taskExecutorAddress);
+ return new
org.apache.flink.runtime.rpc.registration.RegistrationResponse.Decline("cannot
connect to taskExecutor using given address");
+ } else {
+ // save the register taskExecutor
gateway
+ taskExecutorGateways.put(resourceID,
taskExecutorGateway);
+ // schedule the heartbeat with the
registered taskExecutor
+
ResourceManagerToTaskExecutorHeartbeatScheduler heartbeatScheduler = new
ResourceManagerToTaskExecutorHeartbeatScheduler(
+ ResourceManager.this,
leaderSessionID, taskExecutorGateway, taskExecutorAddress, resourceID, log);
+ heartbeatScheduler.start();
+ heartbeatSchedulers.put(resourceID,
heartbeatScheduler);
+ return new
TaskExecutorRegistrationSuccess(new InstanceID(),
heartbeatScheduler.getHeartbeatInterval());
+ }
+ }
+ }, getMainThreadExecutionContext());
+
+ }
+
+ /**
+ * notify resource failure to resourceManager, because of two reasons:
+ * 1. cannot keep heartbeat with taskManager for several times, mark
the resource as failed
+ * 2. in some corner cases, TM will be marked as invalid by cluster
manager master(e.g. yarn master), but TM itself does not realize.
+ *
+ * @param resourceID identify the taskManager which to stop
+ */
+ @RpcMethod
+ public void notifyResourceFailure(ResourceID resourceID) {
+ log.warn("receive failure notification of resource {}",
resourceID);
+ TaskExecutorGateway taskManager =
taskExecutorGateways.get(resourceID);
--- End diff --
Why not calling `taskExecutorGateways.remove(resourceID)` directly here?
---
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.
---
