From: Wolfgang Bumiller <w.bumiller@proxmox.com>
To: Dominik Csapak <d.csapak@proxmox.com>
Cc: "Proxmox VE development discussion" <pve-devel@lists.proxmox.com>,
"Fabian Grünbichler" <f.gruenbichler@proxmox.com>
Subject: Re: [pve-devel] [PATCH qemu-server 1/3] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s
Date: Thu, 22 Sep 2022 14:46:08 +0200 [thread overview]
Message-ID: <20220922124608.hoqg7kbdlhpg52vo@casey.proxmox.com> (raw)
In-Reply-To: <bcf15002-5719-3ba0-fb93-7d0c53a8cb97@proxmox.com>
On Thu, Sep 22, 2022 at 02:22:45PM +0200, Dominik Csapak wrote:
> On 9/22/22 14:01, Wolfgang Bumiller wrote:
> > On Thu, Sep 22, 2022 at 01:31:49PM +0200, Dominik Csapak wrote:
> > > [snip]
> > > > > -/*
> > > > > - * SIGALRM and cleanup handling
> > > > > - *
> > > > > - * terminate_client will set an alarm for 5 seconds and add its client's PID to
> > > > > - * the forced_cleanups list - when the timer expires, we iterate the list and
> > > > > - * attempt to issue SIGKILL to all processes which haven't yet stopped.
> > > > > - */
> > > > > -
> > > > > -static void
> > > > > -alarm_handler(__attribute__((unused)) int signum)
> > > > > -{
> > > > > - alarm_triggered = 1;
> > > > > -}
> > > > > -
> > > >
> > > > wasn't this intentionally decoupled like this?
> > > >
> > > > alarm_handler just sets the flag
> > > > actual force cleanup is conditionalized on the alarm having triggered,
> > > > but the cleanup happens outside of the signal handler..
> > > >
> > > > is there a reason from switching away from these scheme? we don't need
> > > > to do the cleanup in the signal handler (timing is already plenty fuzzy
> > > > anyway ;))
> > >
> > > no real reason, i found the code somewhat cleaner, but you're right,
> > > we probably want to keep that, and just trigger it regularly
> >
> > From what I can tell the only point of this signal is to interrupt
> > `epoll()` after a while to call the cleanup/kill handler since we only
> > have a single worker here that needs to do some work after a timeout.
> >
> > Why not either:
> > - set a bool instead of calling `alarm()` which causes the next
> > `epoll()` call to use a timeout and call the cleanups if epoll turns
> > up empty > - or create a timerfd (timerfd_create(2)) in the beginning which we
> > add to the epoll context and use `timerfd_settime(2)` in place of
> > `alarm()`, which will also wake up the epoll call without having to add
> > timeouts to it
> >
> > `alarm()` is just such a gross interface...
> > In theory we'd also be able to ditch all of those `EINTR` loops as we
> > wouldn't be expecting any interrupts anymore... (and if we did expect
> > them, we could add a `signalfd(2)` to `epoll()` as well ;-)
>
> first one sounds much simpler but the second one sounds much more elegant ;)
> i'll see what works/feels better
>
> couldn't we also directly add a new timerfd for each client that
> needs such a timeout instead of managing some list ?
I'm not really a fan of abusing kernel-side resources for a user-space
scheduling problem ;-).
If you want a per-client timeout, I'd much rather just have a list of
points in time that epoll() should target.
That list may well be `struct Client` itself if you want to merge the
data as you described below. The `struct Client` could just receive a
`STAILQ_ENTRY(Client) cleanup_entries;` member (queue(7), stailq(3)),
and a cleanup time which is used to generate the timeout for `epoll()`,
and on every wakeup, we can march through the already expired queue
entries.
> the cleanupdata could go into the even.data.ptr and we wouldn't
> have to do anything periodically, just handle the timeout
> when epoll wakes up?
>
> we probably would have to merge the client and clenaupdata structs
> so that we can see which is which, but that should not be
> that of a problem?
next prev parent reply other threads:[~2022-09-22 12:46 UTC|newest]
Thread overview: 13+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-09-21 12:49 [pve-devel] [PATCH qemu-server 0/3] qmeventd: improve shutdown behaviour Dominik Csapak
2022-09-21 12:49 ` [pve-devel] [PATCH qemu-server 1/3] qmeventd: rework 'forced_cleanup' handling and set timeout to 60s Dominik Csapak
[not found] ` <<20220921124911.3224970-2-d.csapak@proxmox.com>
2022-09-22 8:24 ` Fabian Grünbichler
2022-09-22 11:31 ` Dominik Csapak
2022-09-22 12:01 ` Wolfgang Bumiller
2022-09-22 12:22 ` Dominik Csapak
2022-09-22 12:46 ` Wolfgang Bumiller [this message]
2022-09-22 11:51 ` Thomas Lamprecht
2022-09-21 12:49 ` [pve-devel] [PATCH qemu-server 2/3] qmeventd: cancel 'forced cleanup' when normal cleanup succeeds Dominik Csapak
2022-09-22 10:14 ` Matthias Heiserer
2022-09-22 11:37 ` Dominik Csapak
2022-09-23 7:58 ` Wolfgang Bumiller
2022-09-21 12:49 ` [pve-devel] [PATCH qemu-server 3/3] qmeventd: send QMP 'quit' command instead of SIGTERM Dominik Csapak
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=20220922124608.hoqg7kbdlhpg52vo@casey.proxmox.com \
--to=w.bumiller@proxmox.com \
--cc=d.csapak@proxmox.com \
--cc=f.gruenbichler@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