[sheepdog] [PATCH v3] sheep: avoid the OOM killer

[Hitoshi Mitake]

At Fri, 5 Jul 2013 14:26:23 +0800,
Liu Yuan wrote:
> 
> On Fri, Jul 05, 2013 at 03:18:48PM +0900, Hitoshi Mitake wrote:
> > Because sheep provides virtual disks to many VMs, sheep is an
> > important process and it shouldn't be killed by the OOM killer of
> > Linux. This patch implements a mechanism for avoiding the OOM killer.
> > 
> > Signed-off-by: Hitoshi Mitake <mitake.hitoshi at lab.ntt.co.jp>
> > ---
> > 
> > v3:
> >  - renaming the function: avoid_oom_killer() -> disable_oom_killing()
> >  - include <linux/oom.h> for the constant OOM_SCORE_ADJ_MIN
> >  - move the function to lib/util.c for other programs like shepherd
> > 
> > v2:
> >  - remove the meaningless comment
> >  - add a description for the value -1000
> > 
> >  include/util.h |    1 +
> >  lib/util.c     |   32 ++++++++++++++++++++++++++++++++
> >  sheep/sheep.c  |    4 ++++
> >  3 files changed, 37 insertions(+)
> > 
> > diff --git a/include/util.h b/include/util.h
> > index e70bb9f..26e4bd9 100644
> > --- a/include/util.h
> > +++ b/include/util.h
> > @@ -102,6 +102,7 @@ void trim_zero_blocks(void *buf, uint64_t *offset, uint32_t *len);
> >  void untrim_zero_blocks(void *buf, uint64_t offset, uint32_t len,
> >  			uint32_t requested_len);
> >  int atomic_create_and_write(const char *path, char *buf, size_t len);
> > +int disable_oom_killing(void);
> >  
> >  /* a type safe version of qsort() */
> >  #define xqsort(base, nmemb, compar)					\
> > diff --git a/lib/util.c b/lib/util.c
> > index 6fd3817..85546dc 100644
> > --- a/lib/util.c
> > +++ b/lib/util.c
> > @@ -23,6 +23,7 @@
> >  #include <signal.h>
> >  #include <sys/xattr.h>
> >  #include <fcntl.h>
> > +#include <linux/oom.h>
> >  
> >  #include "util.h"
> >  #include "logger.h"
> > @@ -539,3 +540,34 @@ close_fd:
> >  end:
> >  	return ret;
> >  }
> > +
> > +int disable_oom_killing(void)
> > +{
> > +	int fd, ret = 0;
> > +	char path[PATH_MAX], score_str[8];
> > +
> > +	snprintf(path, PATH_MAX, "/proc/%d/oom_score_adj", getpid());
> > +	fd = open(path, O_WRONLY);
> > +	if (fd < 0) {
> > +		sd_eprintf("opening %s failed, %m", path);
> > +		ret = -1;
> > +		goto end;
> > +	}
> > +
> > +	snprintf(score_str, 8, "%d\n", OOM_SCORE_ADJ_MIN);
> > +
> > +	/*
> > +	 * -1000 is defined as OOM_SCORE_ADJ_MIN in the linux kernel. And
> > +	 * processes which wrote OOM_SCORE_ADJ_MIN to
> > +	 * /proc/<their pid>/oom_score_adj are excluded by the target list of
> > +	 * the OOM killer certainly.
> > +	 */
> > +	if (xwrite(fd, score_str, strlen(score_str)) != 6) {
> > +		sd_eprintf("writing to %s failed, %m", path);
> > +		ret = -1;
> > +	}
> > +
> > +	close(fd);
> > +end:
> 
> s/end/out

Sorry, I forgot that. I'll fix it in the next version.

Thanks,
Hitoshi