proc.runq.defunct is two times larger than ps shows

[test-account-0]

I can observe that on multiple machines. For example - I have two zombie processes:

root@x8:~# ps -efL H | grep Z
root       6829  200101    6829  0    1 Aug28 ?        Z      0:00 [git] <defunct>
root       97843  394347   97843  0    1  2015 ?        Z      0:00 [sh] <defunct>

and proc.runq.defunct=4.0

Why?

[goodwinos]

looks like that could be a bug - I'll investigate tomorrow if nobody beats me to it

[pcpemail]

We were calling refresh_global_pidlist() twice, resulting in all proc.runq
metrics being doubled. The attached patch fixes it, but may not be correct
for containers. Does anyone know if proc.runq.* metrics are supposed to be
valid for containers? Or are those metrics strictly global?

On Thu, Sep 1, 2016 at 8:51 PM, goodwinos notifications@github.com wrote:

looks like that could be a bug - I'll investigate tomorrow if nobody beats
me to it

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#110 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AL0ItoPTKDHtEBhgYicXE9QeU1X0U9ndks5qlq4XgaJpZM4Jyget
.

---

pcp mailing list
pcp@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/pcp

[pcpemail]

Have posted RH BZ 1372550 "proc.runq.* metric values are double counted"
(this is against RHEL, but the fix will be committed upstream first).

On Fri, Sep 2, 2016 at 1:18 PM, Mark Goodwin mgoodwin@redhat.com wrote:

We were calling refresh_global_pidlist() twice, resulting in all proc.runq
metrics being doubled. The attached patch fixes it, but may not be correct
for containers. Does anyone know if proc.runq.* metrics are supposed to be
valid for containers? Or are those metrics strictly global?

On Thu, Sep 1, 2016 at 8:51 PM, goodwinos notifications@github.com
wrote:

looks like that could be a bug - I'll investigate tomorrow if nobody
beats me to it

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
#110 (comment),
or mute the thread
https://github.com/notifications/unsubscribe-auth/AL0ItoPTKDHtEBhgYicXE9QeU1X0U9ndks5qlq4XgaJpZM4Jyget
.

---

pcp mailing list
pcp@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/pcp

[natoscott]

commit 2fa04a3
Author: Nathan Scott nathans@redhat.com
Date: Thu Sep 8 14:53:49 2016 +1000

pmdaproc: fix double accounting of runq metrics

Fix accidental double accounting of proc runq metrics, and
add missing runq accounting call on the cgroups code path.

Resolves Github Issue #110.
Resolves Red Hat BZ #1372550.