[
https://issues.apache.org/jira/browse/CASSANDRA-15727?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17106575#comment-17106575
]
Benedict Elliott Smith commented on CASSANDRA-15727:
----------------------------------------------------
Sorry that I'm late to this party, but might it be cleaner to move the setting
of {{messagingVersion}} into {{attempt}} since we already do it there, and
remove it from {{onCompletedHandshake/RETRY}} and the class constructor? This
would capture all possible reasons {{messagingVersion}} might have changed, and
reduce the verbosity of this particular point in the code, possibly keeping its
intent a bit clearer.
> Internode messaging connection setup between 4.0 and legacy SSL 3.0 fails if
> initial connection version incorrect
> -----------------------------------------------------------------------------------------------------------------
>
> Key: CASSANDRA-15727
> URL: https://issues.apache.org/jira/browse/CASSANDRA-15727
> Project: Cassandra
> Issue Type: Bug
> Components: Messaging/Internode
> Reporter: Jon Meredith
> Assignee: Jon Meredith
> Priority: Normal
> Labels: pull-request-available
> Fix For: 4.0-alpha
>
> Time Spent: 10m
> Remaining Estimate: 0h
>
> This was discovered while testing upgrading an SSL enabled cluster from 3.0
> to 4.0. The 3.0 cluster was configured to only listen on the ssl storage
> port. When the upgraded 4.0 node started it received a gossip messsage that
> triggered a shadow round before it had correctly set the messaging versions
> for the other endpoints.
> Sending the message created the connection, but because the endpoint
> defaulted to {{VERSION_40}} the initial connect attempt was to the regular
> {{storage_port}}. The 3.0 node was only listening on the
> {{ssl_storage_port}}, so the connection was refused and the
> {{OutboundConnection.onFailure}} handler was called. As the shadow
> gossip round had queued up a message, the {{hasPending}} branch was followed
> and the connection was rescheduled, however the port is never recalculated as
> the original settings are used so it always fails.
> Meanwhile, the node discovered information about peers through inbound
> connection and gossip updating the messaging version for the endpoint which
> could have been used to make a valid connection.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
