[
https://issues.apache.org/jira/browse/CASSANDRA-17752?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17566712#comment-17566712
]
Stefan Miklosovic edited comment on CASSANDRA-17752 at 7/14/22 8:35 AM:
------------------------------------------------------------------------
https://github.com/apache/cassandra/pull/1732
I have renamed this ticket to describe this problem in broader sense. In the
description of the ticket, I am dealing with native transport only.
There was also an idea to get rid of this check completely but I would rather
leave it there for now, preserving the previous solution, I would just fix the
restarting.
The problematic method is also called on these occasions:
1) start of gossip - this is interesting corner case, but I think it should be
possible to stop and start gossipping on a gossipping only member
2) start of thrift - I do not see any reason why restarting of thrift on
gossipping only member should fail
3) start of native transport
4) start of auto compaction - again, this is fairly innocent operation as that
node does not hold any data anyway.
Fixing this will obviously fix respective enable/disable nodetool command
combos.
I will provide all other branches & build once we agree on the approach.
was (Author: smiklosovic):
https://github.com/apache/cassandra/pull/1732
I have renamed this ticket to describe this problem in broader sense. In the
description of the ticket, I am dealing with native transport only.
There was also an idea to get rid of this check completely but I would rather
leave it there for now, preserving the previous solution, I would just fix the
restarting.
The problematic method is also called on these occasions:
1) start of gossip - this is interesting corner case, but I think it should be
possible to stop and start gossipping on a gossipping only member
2) start of thrift - I do not see any reason why restarting of thift on
gossipping only member should fail
3) start of native transport
4) start of auto compaction - again, this is fairly innocent operation as that
node does not hold any data anyway.
Fixing this will obviously fix respective enable/disable nodetool command
combos.
I will provide all other branches & build once we agree on the approach.
> fix restarting of services on gossipping-only members
> -----------------------------------------------------
>
> Key: CASSANDRA-17752
> URL: https://issues.apache.org/jira/browse/CASSANDRA-17752
> Project: Cassandra
> Issue Type: Bug
> Components: Legacy/Core, Tool/nodetool
> Reporter: Stefan Miklosovic
> Assignee: Stefan Miklosovic
> Priority: Normal
> Fix For: 3.0.x, 4.0.x, 4.1.x, 4.x
>
>
> When a node is started with -Dcassandra.join_ring=false, it is possible to
> talk to it via CQL. If we disable it via "nodetool disablebinary", it is not
> possible to enable it back with "nodetool enablebinary".
> The reason why is that enablebinary command eventually calls
> "StorageService#checkServiceAllowedToStart" and it throws when:
> 1) the node is draining (which does not happen)
> 2) it is shutting down (which does not happen)
> 3) is not in NORMAL state - which happens.
> The state of the gossipping-only member is STARTING for ever.
> The check should make an exception here, like if it is in STARTING but
> -Dcassandra.join_ring=false, start it anyway.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
