From: Eric Wong <e@80x24.org>
To: yahns-public@yhbt.net
Subject: [PATCH 2/3] server: workaround Linux v5.5..v5.13 epoll bug
Date: Sat, 9 Oct 2021 02:24:45 +0000 [thread overview]
Message-ID: <20211009022446.705-3-e@80x24.org> (raw)
In-Reply-To: <20211009022446.705-1-e@80x24.org>
epoll_wait() wakeups from QueueQuitter got lost during graceful
shutdown since there's multiple worker threads operating off the
same FD. Workaround the problem by re-arming the eventfd for
every worker thread reaped.
Link: https://yhbt.net/lore/lkml/20210405231025.33829-1-dave@stgolabs.net/
---
lib/yahns/queue_epoll.rb | 4 ++++
lib/yahns/server.rb | 17 ++++++++++-------
2 files changed, 14 insertions(+), 7 deletions(-)
diff --git a/lib/yahns/queue_epoll.rb b/lib/yahns/queue_epoll.rb
index 9e4271a..a198fbf 100644
--- a/lib/yahns/queue_epoll.rb
+++ b/lib/yahns/queue_epoll.rb
@@ -32,6 +32,10 @@ def queue_mod(io, flags)
epoll_ctl(Epoll::CTL_MOD, io, flags)
end
+ def queue_del(io)
+ epoll_ctl(Epoll::CTL_DEL, io, 0)
+ end
+
def thr_init
Thread.current[:yahns_rbuf] = ''.dup
Thread.current[:yahns_fdmap] = @fdmap
diff --git a/lib/yahns/server.rb b/lib/yahns/server.rb
index 208b5ee..74eeb7e 100644
--- a/lib/yahns/server.rb
+++ b/lib/yahns/server.rb
@@ -438,25 +438,28 @@ def quit_enter(alive)
# This just injects the QueueQuitter object which acts like a
# monkey wrench thrown into a perfectly good engine :)
def quit_finish
- quitter = Yahns::QueueQuitter.new
+ # we must not let quitters get GC-ed if we have any worker threads leftover
+ @quitter = Yahns::QueueQuitter.new
# throw the monkey wrench into the worker threads
- @queues.each { |q| q.queue_add(quitter, Yahns::Queue::QEV_QUIT) }
+ @queues.each { |q| q.queue_add(@quitter, Yahns::Queue::QEV_QUIT) }
# watch the monkey wrench destroy all the threads!
# Ugh, this may fail if we have dedicated threads trickling
# response bodies out (e.g. "tail -F") Oh well, have a timeout
begin
@wthr.delete_if { |t| t.join(0.01) }
+ # Workaround Linux 5.5+ bug (fixed in 5.13+)
+ # https://yhbt.net/lore/lkml/20210405231025.33829-1-dave@stgolabs.net/
+ @wthr[0] && @queues[0].respond_to?(:queue_del) and @queues.each do |q|
+ q.queue_del(@quitter)
+ q.queue_add(@quitter, Yahns::Queue::QEV_QUIT)
+ end
end while @wthr[0] && Yahns.now <= @shutdown_expire
# cleanup, our job is done
@queues.each(&:close).clear
-
- # we must not let quitter get GC-ed if we have any worker threads leftover
- @quitter = quitter
-
- quitter.close
+ @quitter.close # keep object around in case @wthr isn't empty
rescue => e
Yahns::Log.exception(@logger, "quit finish", e)
ensure
next prev parent reply other threads:[~2021-10-09 2:24 UTC|newest]
Thread overview: 4+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-09 2:24 [PATCH 0/3] various minor bugfixes Eric Wong
2021-10-09 2:24 ` [PATCH 1/3] do not sleep if signals are pending Eric Wong
2021-10-09 2:24 ` Eric Wong [this message]
2021-10-09 2:24 ` [PATCH 3/3] gemspec: allow unicorn 6.x Eric Wong
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
List information: https://yhbt.net/yahns/README
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20211009022446.705-3-e@80x24.org \
--to=e@80x24.org \
--cc=yahns-public@yhbt.net \
/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.
Code repositories for project(s) associated with this public inbox
https://yhbt.net/yahns.git/
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).