[
https://issues.apache.org/jira/browse/HDDS-16303?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Mohammadhani Fouladgar updated HDDS-16303:
------------------------------------------
Description:
When SCM leadership moves to a newly added SCM that a client (the OM) does not
yet have in its configuration, the client cannot reach the leader and the
operation fails.
*Root cause:*
{{SCMBlockLocationFailoverProxyProvider.performFailoverToAssignedLeader()}}
matches {{ServerNotLeaderException.getSuggestedLeader()}} against
{{scmProxyInfoMap}}; a non-configured leader yields {{updatedLeaderNodeID =
null}} and it round-robins the old followers. The container-location proxy
provider behaves the same way.
*Result:* transferring SCM leadership to a not-yet-known SCM strands every
config-stale SCM client even though the SCM Ratis ring is healthy.
*Proposed change:*
* Allow the SCM failover proxy providers (block-location and
container-location) to construct a proxy to the suggested-leader address
returned in {{ServerNotLeaderException}} even when it is not in static config,
and route to it.
* Do this safely: validate/allowlist the suggested address (require it to
resolve within the configured service/security realm) so a peer cannot redirect
a client to an arbitrary host. This preserves operator-controlled membership
while removing thehard failure.
Acceptance criteria:
* With a new SCM as leader and clients still on the old config, client
operations succeed by following the suggested SCM leader.
* A suggested leader outside the trusted SCM service is rejected (negative
test).
* Applies to both the SCM block-location and container-location failover
providers.
was:When SCM leadership moves to a newly added SCM that a client (the OM)
does not yet have in its configuration, the client cannot reach the leader and
the operation fails.
> Allow SCM clients to route to a Ratis-suggested SCM leader that is not in the
> client's static config (dynamic leader discovery)
> -------------------------------------------------------------------------------------------------------------------------------
>
> Key: HDDS-16303
> URL: https://issues.apache.org/jira/browse/HDDS-16303
> Project: Apache Ozone
> Issue Type: New Feature
> Reporter: Mohammadhani Fouladgar
> Assignee: Mohammadhani Fouladgar
> Priority: Major
>
> When SCM leadership moves to a newly added SCM that a client (the OM) does
> not yet have in its configuration, the client cannot reach the leader and the
> operation fails.
>
> *Root cause:*
> {{SCMBlockLocationFailoverProxyProvider.performFailoverToAssignedLeader()}}
> matches {{ServerNotLeaderException.getSuggestedLeader()}} against
> {{scmProxyInfoMap}}; a non-configured leader yields {{updatedLeaderNodeID =
> null}} and it round-robins the old followers. The container-location proxy
> provider behaves the same way.
> *Result:* transferring SCM leadership to a not-yet-known SCM strands every
> config-stale SCM client even though the SCM Ratis ring is healthy.
> *Proposed change:*
> * Allow the SCM failover proxy providers (block-location and
> container-location) to construct a proxy to the suggested-leader address
> returned in {{ServerNotLeaderException}} even when it is not in static
> config, and route to it.
> * Do this safely: validate/allowlist the suggested address (require it to
> resolve within the configured service/security realm) so a peer cannot
> redirect a client to an arbitrary host. This preserves operator-controlled
> membership while removing thehard failure.
>
> Acceptance criteria:
> * With a new SCM as leader and clients still on the old config, client
> operations succeed by following the suggested SCM leader.
> * A suggested leader outside the trusted SCM service is rejected (negative
> test).
> * Applies to both the SCM block-location and container-location failover
> providers.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
