924060929 commented on code in PR #70693:
URL: https://github.com/apache/doris/pull/70693#discussion_r4628024181
##########
fe/fe-core/src/main/java/org/apache/doris/qe/runtime/PipelineExecutionTaskBuilder.java:
##########
@@ -81,12 +90,13 @@ private Map<BackendWorker, MultiFragmentsPipelineTask>
buildMultiFragmentTasks(
ByteString serializeFragments =
workerToSerializeFragments.get(worker);
Backend backend = worker.getBackend();
+ // Remote catalog workers keep their independent cluster's channel
policy.
fragmentTasks.put(
worker,
new MultiFragmentsPipelineTask(
coordinatorContext,
backend,
- backendServiceProxy,
+ isInternalWorker(worker) ? backendServiceProxy :
BackendServiceProxy.getInstance(false),
Review Comment:
Could we bind the BE target once through `BackendServiceProxy` and let
`MultiFragmentsPipelineTask` and `ResultReceiver` reuse that binding for the
same worker? The coordinator can supply the address and worker's cluster
identity. The RPC layer can then own channel selection, TLS policy, and cache
isolation. Prepare, start, cancel, fetch, and failed-channel eviction would all
use the same target policy, without the builder choosing a proxy here and
`ResultReceiver` independently choosing one from an `internalRpc` flag.
Please keep execution-graph admission in the coordinator before dispatch. A
usable outbound FE-to-BE channel does not tell us whether a remote worker can
report status back to the coordinator or participate in cross-cluster
exchanges. Those checks need the complete worker graph.
--
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.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]