xen-devel.lists.xenproject.org archive mirror
 help / color / mirror / Atom feed
From: Julien Grall <julien@xen.org>
To: Juergen Gross <jgross@suse.com>, xen-devel@lists.xenproject.org
Cc: Ian Jackson <iwj@xenproject.org>, Wei Liu <wl@xen.org>
Subject: Re: [PATCH] tools/xenstore: claim resources when running as daemon
Date: Fri, 14 May 2021 21:19:27 +0100	[thread overview]
Message-ID: <1e38cce0-6960-ac21-b349-dac8551e23ed@xen.org> (raw)
In-Reply-To: <20210514084133.18658-1-jgross@suse.com>

Hi Juergen,

On 14/05/2021 09:41, Juergen Gross wrote:
> Xenstored is absolutely mandatory for a Xen host and it can't be
> restarted, so being killed by OOM-killer in case of memory shortage is
> to be avoided.
> 
> Set /proc/$pid/oom_score_adj (if available) to -500 in order to allow
> xenstored to use large amounts of memory without being killed.
> 
> In order to support large numbers of domains the limit for open file
> descriptors might need to be raised. Each domain needs 2 file
> descriptors (one for the event channel and one for the xl per-domain
> daemon to connect to xenstored).

Hmmm... AFAICT there is only one file descriptor to handle all the event 
channels. Could you point out the code showing one event FD per domain?

> 
> Try to raise ulimit for open files to 65536. First the hard limit if
> needed, and then the soft limit.

I am not sure this is right to impose this limit to everyone. For 
instance, one admin may know that there will be no more than 100 domains 
on its system.

So the admin should be able to configure them. At this point, I think 
the two limit should be set my the initscript rather than xenstored itself.

This would also avoid the problem where Xenstored is not allowed to 
modify its limit (see more below).

> 
> Signed-off-by: Juergen Gross <jgross@suse.com>
> ---
>   tools/xenstore/xenstored_core.c   |  2 ++
>   tools/xenstore/xenstored_core.h   |  3 ++
>   tools/xenstore/xenstored_minios.c |  4 +++
>   tools/xenstore/xenstored_posix.c  | 46 +++++++++++++++++++++++++++++++
>   4 files changed, 55 insertions(+)
> 
> diff --git a/tools/xenstore/xenstored_core.c b/tools/xenstore/xenstored_core.c
> index b66d119a98..964e693450 100644
> --- a/tools/xenstore/xenstored_core.c
> +++ b/tools/xenstore/xenstored_core.c
> @@ -2243,6 +2243,8 @@ int main(int argc, char *argv[])
>   		xprintf = trace;
>   #endif
>   
> +	claim_resources();
> +
>   	signal(SIGHUP, trigger_reopen_log);
>   	if (tracefile)
>   		tracefile = talloc_strdup(NULL, tracefile);
> diff --git a/tools/xenstore/xenstored_core.h b/tools/xenstore/xenstored_core.h
> index 1467270476..ac26973648 100644
> --- a/tools/xenstore/xenstored_core.h
> +++ b/tools/xenstore/xenstored_core.h
> @@ -255,6 +255,9 @@ void daemonize(void);
>   /* Close stdin/stdout/stderr to complete daemonize */
>   void finish_daemonize(void);
>   
> +/* Set OOM-killer score and raise ulimit. */
> +void claim_resources(void);
> +
>   /* Open a pipe for signal handling */
>   void init_pipe(int reopen_log_pipe[2]);
>   
> diff --git a/tools/xenstore/xenstored_minios.c b/tools/xenstore/xenstored_minios.c
> index c94493e52a..df8ff580b0 100644
> --- a/tools/xenstore/xenstored_minios.c
> +++ b/tools/xenstore/xenstored_minios.c
> @@ -32,6 +32,10 @@ void finish_daemonize(void)
>   {
>   }
>   
> +void claim_resources(void)
> +{
> +}
> +
>   void init_pipe(int reopen_log_pipe[2])
>   {
>   	reopen_log_pipe[0] = -1;
> diff --git a/tools/xenstore/xenstored_posix.c b/tools/xenstore/xenstored_posix.c
> index 48c37ffe3e..0074fbd8b2 100644
> --- a/tools/xenstore/xenstored_posix.c
> +++ b/tools/xenstore/xenstored_posix.c
> @@ -22,6 +22,7 @@
>   #include <fcntl.h>
>   #include <stdlib.h>
>   #include <sys/mman.h>
> +#include <sys/resource.h>
>   
>   #include "utils.h"
>   #include "xenstored_core.h"
> @@ -87,6 +88,51 @@ void finish_daemonize(void)
>   	close(devnull);
>   }
>   
> +static void avoid_oom_killer(void)
> +{
> +	char path[32];
> +	char val[] = "-500";
> +	int fd;
> +
> +	snprintf(path, sizeof(path), "/proc/%d/oom_score_adj", (int)getpid());

This looks Linux specific. How about other OSes?

> +
> +	fd = open(path, O_WRONLY);
> +	/* Do nothing if file doesn't exist. */

Your commit message leads to think that we *must* configure the OOM. If 
not, then we should not continue. But here, this suggest this is 
optional. In fact...

> +	if (fd < 0)
> +		return;
> +	/* Ignore errors. */
> +	write(fd, val, sizeof(val));

... xenstored may not be allowed to modify its own parameters. So this 
would continue silently without the admin necessarily knowning the limit 
wasn't applied.

> +	close(fd);
> +}
> +
> +/* Max. 32752 domains with 2 open files per domain, plus some spare. */
> +#define MAX_FILES 65536
> +static void raise_ulimit(void)
> +{
> +	struct rlimit rlim;
> +
> +	if (getrlimit(RLIMIT_NOFILE, &rlim))
> +		return;
> +	if (rlim.rlim_max != RLIM_INFINITY && rlim.rlim_max < MAX_FILES)
> +	{
> +		rlim.rlim_max = MAX_FILES;
> +		setrlimit(RLIMIT_NOFILE, &rlim);
> +	}
> +	if (getrlimit(RLIMIT_NOFILE, &rlim))
> +		return;
> +	if (rlim.rlim_max == RLIM_INFINITY || rlim.rlim_max > MAX_FILES)
> +		rlim.rlim_cur = MAX_FILES;
> +	else
> +		rlim.rlim_cur = rlim.rlim_max;
> +	setrlimit(RLIMIT_NOFILE, &rlim);
> +}
> +
> +void claim_resources(void)
> +{
> +	avoid_oom_killer();
> +	raise_ulimit();
> +}
> +
>   void init_pipe(int reopen_log_pipe[2])
>   {
>   	int flags;
> 

Cheers,

-- 
Julien Grall


  reply	other threads:[~2021-05-14 20:19 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-05-14  8:41 Juergen Gross
2021-05-14 20:19 ` Julien Grall [this message]
2021-05-17  6:47   ` Juergen Gross
2021-05-17 15:55     ` Julien Grall
2021-05-18  6:43       ` Juergen Gross
2021-05-18 13:09         ` Julien Grall

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=1e38cce0-6960-ac21-b349-dac8551e23ed@xen.org \
    --to=julien@xen.org \
    --cc=iwj@xenproject.org \
    --cc=jgross@suse.com \
    --cc=wl@xen.org \
    --cc=xen-devel@lists.xenproject.org \
    --subject='Re: [PATCH] tools/xenstore: claim resources when running as daemon' \
    /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

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 NNTP newsgroup(s).