[
https://issues.apache.org/jira/browse/HDDS-16335?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=18110904#comment-18110904
]
Chia-Chuan Ho commented on HDDS-16335:
--------------------------------------
Hi [~szetszwo]
I took an initial look at XceiverClientShortCircuit and would like to check my
understanding before starting the change.
Would it make sense to use a single lock to coordinate connection
initialization, the open check and request write, and the close transition? The
receiver and callers waiting for responses would remain outside the lock. This
should still allow multiple requests to be in flight, while keeping request
frames serialized and preventing close() from racing with socket writes and
timeout registration.
My current understanding is that this client is intended to use a one-shot
connection, since the receiver daemon cannot be restarted. I would therefore
keep repeated connect() idempotent while the connection is open, and reject
reconnect attempts after failure or close.
I am also considering making the request/response counters safe for
cross-thread access and letting the receiver use stable references to the
socket and timer. For coverage, I would extend the existing short-circuit
integration test with concurrent connect(), Echo requests using distinct call
IDs, and concurrent close().
I noticed a related issue around exact-once cleanup and pending metrics when
the connection fails. To keep this patch focused, I was thinking of handling
that separately, unless you think it belongs in this Jira.
Does this approach sound reasonable, or is there another synchronization model
you had in mind?
Thanks!
> Change XceiverClientShortCircuit to support concurrent access
> -------------------------------------------------------------
>
> Key: HDDS-16335
> URL: https://issues.apache.org/jira/browse/HDDS-16335
> Project: Apache Ozone
> Issue Type: Sub-task
> Components: Ozone Client
> Reporter: Tsz-wo Sze
> Priority: Major
>
> In XceiverClientShortCircuit, domainSocket and some other fields are not
> synchronized.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
