Can pid be reused ?

[Loic Dachary]

Hi,

Something strange happens on fedora20 with linux 3.11.10-301.fc20.x86_64. Running make -j8 check on https://github.com/ceph/ceph/pull/2750 a process gets killed from time to time. For instance it shows as

TEST_erasure_crush_stripe_width: 124: stripe_width=4096
TEST_erasure_crush_stripe_width: 125: ./ceph osd pool create pool_erasure 12 12 erasure
*** DEVELOPER MODE: setting PATH, PYTHONPATH and LD_LIBRARY_PATH ***
./test/mon/osd-pool-create.sh: line 120: 27557 Killed ./ceph osd pool create pool_erasure 12 12 erasure
TEST_erasure_crush_stripe_width: 126: ./ceph --format json osd dump
TEST_erasure_crush_stripe_width: 126: tee osd-pool-create/osd.json

in the test logs. Note the 27557 Killed . I originally thought it was because some ulimit was crossed and set them to very generous / unlimited hard / soft thresholds.

core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 515069
max locked memory (kbytes, -l) unlimited
max memory size (kbytes, -m) unlimited
open files (-n) 400000
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) unlimited
cpu time (seconds, -t) unlimited
max user processes (-u) unlimited
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited

Benoit Canet suggested that I installed systemtap ( https://www.sourceware.org/systemtap/wiki/SystemtapOnFedora ) and ran https://sourceware.org/systemtap/examples/process/sigkill.stp to watch what was sending the kill signal. It showed the following:

...
SIGKILL was sent to ceph-osd (pid:27557) by vstart_wrapper. uid:1001
SIGKILL was sent to python (pid:27557) by vstart_wrapper. uid:1001
....

which suggests that pid 27557 used by ceph-osd was reused for the python script that was killed above. Because the script that kills daemons is very agressive and kill -9 the pid to check if it really is dead

https://github.com/ceph/ceph/blob/giant/src/test/mon/mon-test-helpers.sh#L64

it explains the problem.

However, as Dan Mick suggests, reusing pid quickly could break a number of things and it is a surprising behavior. Maybe something else is going on. A loop creating processes sees their pid increasing and not being reused.

Any idea about what is going on would be much appreciated :-)

Cheers

[David Zafman]

I just realized what it is. The way killall is used when stopping a vs=
tart cluster, is to kill all processes by name! You can't stop vstarte=
d tests running in parallel.

David Zafman
Senior Developer
http://www.inktank.com

=20
Hi,
=20
Something strange happens on fedora20 with linux 3.11.10-301.fc20.x86=

_64. Running make -j8 check on https://github.com/ceph/ceph/pull/2750 a=
process gets killed from time to time. For instance it shows as

=20
TEST_erasure_crush_stripe_width: 124: stripe_width=3D4096
TEST_erasure_crush_stripe_width: 125: ./ceph osd pool create pool_era=

sure 12 12 erasure

*** DEVELOPER MODE: setting PATH, PYTHONPATH and LD_LIBRARY_PATH ***
./test/mon/osd-pool-create.sh: line 120: 27557 Killed =

./ceph osd pool create pool_erasure 12 12 erasure

TEST_erasure_crush_stripe_width: 126: ./ceph --format json osd dump
TEST_erasure_crush_stripe_width: 126: tee osd-pool-create/osd.json
=20
in the test logs. Note the 27557 Killed . I originally thought it was=

because some ulimit was crossed and set them to very generous / unlimi=
ted hard / soft thresholds.

=20
core file size (blocks, -c) 0 =

=20

data seg size (kbytes, -d) unlimited =

=20

scheduling priority (-e) 0 =

=20

file size (blocks, -f) unlimited =

=20

pending signals (-i) 515069 =

=20

max locked memory (kbytes, -l) unlimited =

=20

max memory size (kbytes, -m) unlimited =

=20

open files (-n) 400000 =

=20

pipe size (512 bytes, -p) 8 =

=20

POSIX message queues (bytes, -q) 819200 =

=20

real-time priority (-r) 0 =

=20

stack size (kbytes, -s) unlimited =

=20

cpu time (seconds, -t) unlimited =

=20

max user processes (-u) unlimited =

=20

virtual memory (kbytes, -v) unlimited =

=20

file locks (-x) unlimited =20
=20
Benoit Canet suggested that I installed systemtap ( https://www.sourc=

eware.org/systemtap/wiki/SystemtapOnFedora ) and ran https://sourceware=
=2Eorg/systemtap/examples/process/sigkill.stp to watch what was sending=

=20
...
SIGKILL was sent to ceph-osd (pid:27557) by vstart_wrapper. uid:1001
SIGKILL was sent to python (pid:27557) by vstart_wrapper. uid:1001
....
=20
which suggests that pid 27557 used by ceph-osd was reused for the pyt=

hon script that was killed above. Because the script that kills daemons=
is very agressive and kill -9 the pid to check if it really is dead

=20
https://github.com/ceph/ceph/blob/giant/src/test/mon/mon-test-helpers=

=2Esh#L64

=20
it explains the problem.
=20
However, as Dan Mick suggests, reusing pid quickly could break a numb=

er of things and it is a surprising behavior. Maybe something else is g=
oing on. A loop creating processes sees their pid increasing and not be=
ing reused.

=20
Any idea about what is going on would be much appreciated :-)
=20
Cheers
=20
--=20
Lo=EFc Dachary, Artisan Logiciel Libre
=20
=20

--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" i=
n
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

[Sage Weil]

Post by David Zafman
I just realized what it is. The way killall is used when stopping a
vstart cluster, is to kill all processes by name! You can't stop
vstarted tests running in parallel.

Ah. FWIW I think we should avoid using stop.sh whenever possible and
instead do ./init-ceph stop (which does an orderly shutdown via pid
files).

sage
--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

[David Zafman]

=20

I just realized what it is. The way killall is used when stopping a=

=20

vstart cluster, is to kill all processes by name! You can't stop=20
vstarted tests running in parallel.

=20
Ah. FWIW I think we should avoid using stop.sh whenever possible and=

=20

instead do ./init-ceph stop (which does an orderly shutdown via pid=20
files).
=20
sage

Actually, vstart.sh can=E2=80=99t create 2 independent clusters anyway,=
so it kills any existing processes. Probably vstart.sh is what would =
have killed the processes in a parallel make check.

David--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" i=
n
the body of a message to ***@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

[Loic Dachary]

Post by Sage Weil

Post by David Zafman
I just realized what it is. The way killall is used when stopping a
vstart cluster, is to kill all processes by name! You can't stop
vstarted tests running in parallel.

Ah. FWIW I think we should avoid using stop.sh whenever possible and
instead do ./init-ceph stop (which does an orderly shutdown via pid
files).
sage

Actually, vstart.sh can’t create 2 independent clusters anyway, so it kills any existing processes.

It can actually, if given a different CEPH_DIR all is contained within this specific directory.

Cheers

Probably vstart.sh is what would have killed the processes in a parallel make check.
David

[Loic Dachary]

Hi David,

On 22/10/2014 15:21, David Zafman wrote:>

I just realized what it is. The way killall is used when stopping a vstart cluster, is to kill all processes by name! You can't stop vstarted tests running in parallel.

I discovered this indeed. But then instead of using ./stop.sh I use

https://github.com/dachary/ceph/blob/6e6ddfbdc0a178a6318a86fd9984265bbe40ca3d/src/test/mon/mon-test-helpers.sh#L62

in the context of

https://github.com/dachary/ceph/blob/6e6ddfbdc0a178a6318a86fd9984265bbe40ca3d/src/test/vstart_wrapper.sh#L28

which makes it kill only the processes with a pid file in the relevant directory. The problem bellow showed because it was doing an aggressive kill -9 to check if the process still exists.

https://github.com/dachary/ceph/commit/6e6ddfbdc0a178a6318a86fd9984265bbe40ca3d

Now that it's replaced with a kill -0 all is well.

For the record the problem can be reliably reproduced by running make -j8 check from https://github.com/dachary/ceph/commit/c02bb8a5afef8669005c78b2b4f2f762cda4ee73 and waiting less than one hour and probably more than 30 minutes on a 24 core, 64GB RAM, 250GB SSD disk.

Cheers

David Zafman
Senior Developer
http://www.inktank.com

Post by Loic Dachary
Hi,
Something strange happens on fedora20 with linux 3.11.10-301.fc20.x86_64. Running make -j8 check on https://github.com/ceph/ceph/pull/2750 a process gets killed from time to time. For instance it shows as
TEST_erasure_crush_stripe_width: 124: stripe_width=4096
TEST_erasure_crush_stripe_width: 125: ./ceph osd pool create pool_erasure 12 12 erasure
*** DEVELOPER MODE: setting PATH, PYTHONPATH and LD_LIBRARY_PATH ***
./test/mon/osd-pool-create.sh: line 120: 27557 Killed ./ceph osd pool create pool_erasure 12 12 erasure
TEST_erasure_crush_stripe_width: 126: ./ceph --format json osd dump
TEST_erasure_crush_stripe_width: 126: tee osd-pool-create/osd.json
in the test logs. Note the 27557 Killed . I originally thought it was because some ulimit was crossed and set them to very generous / unlimited hard / soft thresholds.
core file size (blocks, -c) 0
data seg size (kbytes, -d) unlimited
scheduling priority (-e) 0
file size (blocks, -f) unlimited
pending signals (-i) 515069
max locked memory (kbytes, -l) unlimited
max memory size (kbytes, -m) unlimited
open files (-n) 400000
pipe size (512 bytes, -p) 8
POSIX message queues (bytes, -q) 819200
real-time priority (-r) 0
stack size (kbytes, -s) unlimited
cpu time (seconds, -t) unlimited
max user processes (-u) unlimited
virtual memory (kbytes, -v) unlimited
file locks (-x) unlimited
...
SIGKILL was sent to ceph-osd (pid:27557) by vstart_wrapper. uid:1001
SIGKILL was sent to python (pid:27557) by vstart_wrapper. uid:1001
....
which suggests that pid 27557 used by ceph-osd was reused for the python script that was killed above. Because the script that kills daemons is very agressive and kill -9 the pid to check if it really is dead
https://github.com/ceph/ceph/blob/giant/src/test/mon/mon-test-helpers.sh#L64
it explains the problem.
However, as Dan Mick suggests, reusing pid quickly could break a number of things and it is a surprising behavior. Maybe something else is going on. A loop creating processes sees their pid increasing and not being reused.
Any idea about what is going on would be much appreciated :-)
Cheers
--
Loïc Dachary, Artisan Logiciel Libre

--
To unsubscribe from this list: send the line "unsubscribe ceph-devel" in
More majordomo info at http://vger.kernel.org/majordomo-info.html