Support for query heartbeat from coordinator that is shutting down #16322
Conversation
abhiseksaikia
force-pushed
the
asaikia_coord_drain
branch
from
125335f
to
02155d5
Compare
| @@ -141,8 +144,10 @@ public void registerQueryHeartbeat(String nodeId, BasicQueryInfo basicQueryInfo) | |||
| { | |||
| requireNonNull(nodeId, "nodeId is null"); | |||
| requireNonNull(basicQueryInfo, "basicQueryInfo is null"); | |||
| Stream<InternalNode> activeOrShuttingDownCoordinators = concat(internalNodeManager.getCoordinators().stream(), | |||
| internalNodeManager.getNodes(SHUTTING_DOWN).stream().filter(InternalNode::isCoordinator)); | |||
There was a problem hiding this comment.
it might make sense to expose Shutting down coordinator as a seperate method from InternalNodeManager. i.e. 'getShuttingDownCoordinator` or something similar.
There was a problem hiding this comment.
Also wondering, what value are we going to get with this check here where a coordinator sends it's heartbeat to the RM and we check if it belongs to the same set of nodes or not? We don't do same for the node heartbeat. @tdcmeehan Any specific reason for this check here or can we remove this entirely?
There was a problem hiding this comment.
I don't see any other reason other than a scenario where RM receives a heart beat from a coordinator that does not belong to the cluster, not sure if this is possible. I will wait for Tim to comment.
There was a problem hiding this comment.
It's more of a check to ensure that we're not making decisions based on heartbeats which are inconsistent with discovery service. I would like to eventually add a similar check for node heartbeats.
abhiseksaikia
force-pushed
the
asaikia_coord_drain
branch
2 times, most recently
from
02be719
to
7bc2f97
Compare
Currently query heartbeat is failing to register if it comes from a coordinator node which is shutting down. The fix here is to consider query heartbeat from a shutting down coordinator as a valid one.
abhiseksaikia
force-pushed
the
asaikia_coord_drain
branch
from
7bc2f97
to
a1cd0b7
Compare
Currently query heartbeat is failing to register if it comes from a coordinator node which is shutting down. The fix here is to consider query heartbeat from a shutting down coordinator as a valid one.
Test plan - unit test