Michael Ho created IMPALA-9006:
----------------------------------
Summary: Consolidate the Statestore subscriber's retry logic
Key: IMPALA-9006
URL: https://issues.apache.org/jira/browse/IMPALA-9006
Project: IMPALA
Issue Type: Improvement
Components: Distributed Exec
Affects Versions: Impala 3.4.0
Reporter: Michael Ho
Assignee: Michael Ho
Currently, a Statestore subscriber starts a separate thread after the initial
registration with Statestore to periodically check if the Statestore may have
failed and re-registered with Statestore if necessary. Similarly, the function
{{StatestoreSubscriber::Register()}} also relies on the old Thrift client's
retry logic to retry failed RPC attempts to Statestore. This is needed as the
initial registration relies on this retry logic to wait for Statestore to
startup in case an Impala daemon starts before the Statestore.
Last but not least, the current registration logic at Statestore doesn't check
if the address provided by the subscriber can actually be resolved. In certain
deployment scenarios, it's possible that the address passed by a subscriber is
not yet resolvable (e.g. a Kubernetes pod whose readiness probe failed).
Statestore should check for if the address is resolvable and fail the
registration if not. The subscriber can keep retrying until its address can be
resolved by Statestore. This is particularly useful in configuration where
readiness probe of a pod in Kubernetes is exploited for a warm backup
configuration.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
