smiklosovic commented on code in PR #4394:
URL: https://github.com/apache/cassandra/pull/4394#discussion_r2377826895
##########
src/java/org/apache/cassandra/gms/Gossiper.java:
##########
@@ -2070,6 +2071,74 @@ public void setLooseEmptyEnabled(boolean enabled)
EndpointState.LOOSE_DEF_OF_EMPTY_ENABLED = enabled;
}
+ @Override
+ public Map<String, Map<String, Object>> echoAllNodesWithTiming()
Review Comment:
Thank you for your draft! I am aware that this is a draft and not a proper
PR yet but I think it is OK to comment on this already.
My conceptual issue with this is that, imagine you have a cluster of 200
nodes and you execute this JMX method. If you have e.g. a cluster of 5 nodes,
even 10, it does not look like a big issue. But if you have 100+, then it is.
You have specified that the timeout is 5 seconds, right? So 200x5 = 1000s ~
16 min. So _at worst_, we would need to wait 16 minutes until this method
returns? What happens if you "ctrl+c" nodetool ping? What happens with the
echoes being scheduled for the execution?
To speed up the pinging, you might either ping multiple nodes at once (e.g.
bulk pinging, 10 nodes simultaneously so it happens in parallel), We do not
want to ping "one node at a time" because it is too slow and we do not want to
ping "everybody at once" as it would flood the cluster with echoes.
Simultaneous pinging for predefined number of nodes (might be configurable
by nodetool parameter) is a compromise between "waiting too long" and not
flooding cluster with echoes.
It is still questionable what would happen if we kill pining nodetool
command before it returns. We might schedule a "pinging task" and return
immediately and the result would be stored. Then we might query the result
independently. That way we might also see if pining is still in progress or
finished etc.
Also, it would be nice to have a way to specify DC I want to ping and also
rack in specific DC I want to ping. You can add that as flags to nodetool
command.
--
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]
