Update RemoteClusterService skip_unavailable handling for CPS #132478
Conversation
JeremyDahlgren
added
Team:Distributed Coordination
In stateless CPS skip_unavailable will not be used and should always be true. Also DisconnectedStrategy.RECONNECT_UNLESS_SKIP_UNAVAILABLE is rejected in CPS as a parameter value for getRemoteClusterClient(). Resolves: ES-12267
JeremyDahlgren
force-pushed
the
tasks/es-12267-skip-unavailable-for-cps
branch
from
5510b9d to
c1cd712
Compare
|
Pinging @elastic/es-distributed-coordination (Team:Distributed Coordination) |
server/src/main/java/org/elasticsearch/transport/RemoteClusterService.java
Outdated
Show resolved
Hide resolved
server/src/main/java/org/elasticsearch/transport/RemoteClusterService.java
Outdated
Show resolved
Hide resolved
server/src/main/java/org/elasticsearch/transport/RemoteClusterService.java
Outdated
Show resolved
Hide resolved
server/src/main/java/org/elasticsearch/transport/RemoteClusterService.java
Outdated
Show resolved
Hide resolved
|
I'll need to split this into separate PRs. We can make the skip_unavailable settings change now, but the client call change for |
elasticsearchmachine
added
the
serverless-linked
server/src/main/java/org/elasticsearch/transport/RemoteClusterService.java
Outdated
Show resolved
Hide resolved
server/src/test/java/org/elasticsearch/transport/RemoteClusterSettingsTests.java
Outdated
Show resolved
Hide resolved
server/src/test/java/org/elasticsearch/transport/RemoteClusterSettingsTests.java
Outdated
Show resolved
Hide resolved
| } | ||
| }; | ||
|
|
||
| private static class UnsupportedInStatelessValidator<T> extends RemoteConnectionEnabled<T> { |
There was a problem hiding this comment.
Should we use "serverless" rather than "stateless"? We typically use the terms "stateful" vs. "serverless", so when I see "stateless" I think have to take a moment to parse what it means since the first half of the word feels like one and the second half like the other.
There was a problem hiding this comment.
Oh, I see why you used it - because of the pre-existing DiscoveryNode.STATELESS_ENABLED_SETTING_NAME. In that case, I guess it makes sense to keep using "stateless" here.
There was a problem hiding this comment.
Switch to CPS in the name since the RemoteClusterService and related code all now test the CPS specific setting.
JeremyDahlgren
removed
the
serverless-linked
4 participants
In serverless cross-project search (CPS)
skip_unavailablewill not be used and should always betrue. This PR adds a settings validator that rejects setting skip_unavailable when CPS is enabled.
Resolves: ES-12267