Re: deploying both NFS client and server on the same machine trigger hungtask

[Chuck Lever via Bugspray Bot <bugbot@kernel.org>]

Chuck Lever writes via Kernel.org Bugzilla:

(In reply to Bugspray Bot from comment #1)
> Jeff Layton <jlayton@kernel.org> writes:
> 
> On Tue, 2024-12-03 at 02:30 +0000, Li Lingfeng via Bugspray Bot wrote:
> > Li Lingfeng writes via Kernel.org Bugzilla:
> > 
> > We deployed the client and server on the same machine for some testing,
> which
> > caused some processes to trigger hungtask.

Thanks for the report.


> Note that this is really an unsupported configuration. It's very
> difficult to prevent a situation where the server needs memory in order
> to handle an RPC from the local client, and then tries to write back
> dirty NFS client pages in order to get it. Instant deadlock.
> 
> That kind of configuration is fine for basic functionality testing, but
> once you start getting into memory pressure, it can fall over pretty
> quickly.

Historically, yes, this configuration is officially unsupported. We treat this configuration as best effort only because it is not possible to close all possible writeback deadlocks, as you said above.

However, Neil has mentioned that it is an important usage scenario for one or more of SuSE's customers. And, it is quite broadly deployed for automated testing.

This particular bug appears to be locking related, not writeback related. IMO, deeper investigation is warranted.

View: https://bugzilla.kernel.org/show_bug.cgi?id=219550#c2
You can reply to this message to join the discussion.
-- 
Deet-doot-dot, I am a bot.
Kernel.org Bugzilla (bugspray 0.1-dev)