public inbox for pve-devel@lists.proxmox.com
 help / color / mirror / Atom feed
From: Kefu Chai <k.chai@proxmox.com>
To: pve-devel@lists.proxmox.com
Subject: [PATCH v3 http-server 0/1] fix pveproxy OOM in websocket and spice proxy handlers
Date: Fri, 24 Apr 2026 20:11:39 +0800	[thread overview]
Message-ID: <20260424121140.3687865-1-k.chai@proxmox.com> (raw)

see v2's cover letter [1] for the problem description and the approach.

Changes since v2:

* extract handle_proxy_eof(); the four on_eof sites were copy-paste of
  each other with only $reader and the peer handle differing.

* fix a busy-loop in the on_eof drain loop: v2's unguarded
  `while length($hdl->{rbuf})` spins when the reader's
  `return if !$peer` short-circuits without consuming rbuf. reachable
  on a ws client close that sets block_disconnect on the backend
  handle, so a final reply from the backend pins the worker at 100%
  CPU instead of completing teardown. the new loop bails on
  peer-gone or zero progress.

* clear on_drain in apply_read_backpressure() after firing instead of
  leaving the wrapper installed when prev_on_drain is undef. no
  functional impact (idempotent re-set of on_read) but stops pinning
  a reader reference for the rest of the connection.

both of the above are verified with the same synthetic AnyEvent setup
used for v1/v2. reverting just the busy-loop guard reproduces a spin
that trips a 2 s alarm; reverting just the on_drain clear leaves the
wrapper installed after the drain.

on the peer-gone branch the drain loop no-ops and rbuf is released on
handle teardown, same as the pre-v2 behavior (before this series added
on_eof draining, rbuf at on_eof was always discarded). I audited the
users:

* PDM migration's control tunnel (mtunnel) completes each command
  synchronously via write_tunnel, so its teardown carries no protocol
  data; disk data goes over separate NBD-over-ws tunnels set up by
  forward_unix_socket, and a connection drop there surfaces as a clean
  migration abort on the source side rather than silent corruption.
* NoVNC and SPICE display (plus termproxy shell output) lose at most a
  final frame or line, cosmetic.
* SPICE USB passthrough is the one case with potential real data loss,
  but that requires an abrupt ws client close mid-transfer, which is rare.

[1] https://lore.proxmox.com/pve-devel/20260413125650.2569621-1-k.chai@proxmox.com/

Kefu Chai (1):
  fix #7483: apiserver: add backpressure to proxy handlers

 src/PVE/APIServer/AnyEvent.pm | 178 +++++++++++++++++++++++++---------
 1 file changed, 133 insertions(+), 45 deletions(-)

-- 
2.47.3





             reply	other threads:[~2026-04-24 12:19 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-04-24 12:11 Kefu Chai [this message]
2026-04-24 12:11 ` [PATCH v3 http-server 1/1] fix #7483: apiserver: add backpressure to " Kefu Chai

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260424121140.3687865-1-k.chai@proxmox.com \
    --to=k.chai@proxmox.com \
    --cc=pve-devel@lists.proxmox.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
Service provided by Proxmox Server Solutions GmbH | Privacy | Legal