3 vmstat procs and other issues 🔗 link

On 1/5/2017 1:40 PM, Galen Johnson wrote:

Hey,

While I'm questioning things, I noticed that there are 3 vmstat calls run on the server every 5 minutes instead of the 1 that I would expect...Anyone else seeing that behavior?

I'm running the Terabithia RPM (yes, I know there is a new release about to come out) on Centos 7.

wrote:

On 1/5/2017 1:40 PM, Galen Johnson wrote:

Hey,

While I'm questioning things, I noticed that there are 3 vmstat calls
run on the server every 5 minutes instead of the 1 that I would
expect...Anyone else seeing that behavior?

I'm running the Terabithia RPM (yes, I know there is a new release about
to come out) on Centos 7.

This is actually normal. It's a side effect of the fact that those RPMs
fire the client off every 100s instead of every 5m. Although vmstat (and
anything launched the same way) is collecting info for the previous 5m,
it's doing it once for each client execution that occurs during that time
(3x). You end up with a rolling 5m average than a discrete 5m block when
$interval != $collectionperiod.

I'm also not getting the disk column and most of the other local tests

that I've come to expect, either (memory, cpu, files, etc).

FQDN issue possible here as well?

-jc

=G=

On Thu, Jan 5, 2017 at 6:06 PM, Galen Johnson <user-fc632e705d24@xymon.invalid> wrote:

Ok...I fixed this by adding '-x tmpfs -x devtmpfs' to the df command.
Definitely something you may want to consider.  It was cluttering up my
disk graphs badly (especially on systems that had lots of users logged
in).

=G=

On Thu, Jan 5, 2017 at 6:00 PM, Galen Johnson <user-fc632e705d24@xymon.invalid> wrote:

Thanks for the explanation on the vmstat point...I feel much better now,
I hadn't actually noticed the differences in the timestamps until you
pointed it out.

The FQDN was the issue with my missing tests.  They show up now.  Now I
just need the df command to ignore tmpfs paths an I think I'll be good.

=G=

On Thu, Jan 5, 2017 at 5:15 PM, Japheth Cleaver <user-87556346d4af@xymon.invalid>
wrote:

On 1/5/2017 1:40 PM, Galen Johnson wrote:

Hey,

While I'm questioning things, I noticed that there are 3 vmstat calls
run on the server every 5 minutes instead of the 1 that I would
expect...Anyone else seeing that behavior?

I'm running the Terabithia RPM (yes, I know there is a new release
about to come out) on Centos 7.

This is actually normal. It's a side effect of the fact that those RPMs
fire the client off every 100s instead of every 5m. Although vmstat (and
anything launched the same way) is collecting info for the previous 5m,
it's doing it once for each client execution that occurs during that time
(3x). You end up with a rolling 5m average than a discrete 5m block when
$interval != $collectionperiod.

I'm also not getting the disk column and most of the other local tests

that I've come to expect, either (memory, cpu, files, etc).

FQDN issue possible here as well?

-jc

On 1/5/2017 3:18 PM, Galen Johnson wrote:

Ok...looking more closely at the xymonclient-linux.sh script, it appears that tmpfs was intentionally not included in the exclude list.  At the very least, you may want to consider adding /run to the default excludes in analysis.cfg. I was also noticing in my df output that other virtual filesystems were showing (and being tracked) as well, such as /dev/shm and /sys.  I'm not sure why you would want those tracked either.  For example, here's a list of tmpfs "disks" I see on one of my systems:

tmpfs                 2033585      9   2033576    1% /dev/shm
tmpfs                 2033585    598   2032987    1% /run
tmpfs                 2033585     13   2033572    1% /sys/fs/cgroup
tmpfs                 2033585     29   2033556    1% /tmp
tmpfs                 2033585      1   2033584    1% /run/user/1984

Way back on 6 Jan 2017, at 3:00, Japeth Cleaver <user-87556346d4af@xymon.invalid> wrote:

On 1/5/2017 3:18 PM, Galen Johnson wrote:

Ok...looking more closely at the xymonclient-linux.sh script, it
appears that tmpfs was intentionally not included in the exclude
list.  At the very least, you may want to consider adding /run to the
default excludes in analysis.cfg.  I was also noticing in my df output
that other virtual filesystems were showing (and being tracked) as
well, such as /dev/shm and /sys.  I'm not sure why you would want
those tracked either.  For example, here's a list of tmpfs "disks" I
see on one of my systems:

tmpfs                 2033585      9   2033576    1% /dev/shm
tmpfs                 2033585    598   2032987    1% /run
tmpfs                 2033585     13   2033572    1% /sys/fs/cgroup
tmpfs                 2033585     29   2033556    1% /tmp
tmpfs                 2033585      1   2033584    1% /run/user/1984

Oy.

I'd forgotten about the newer per-user /run/ environments that were
coming in now in EL7.  Definitely worth adding specific excludes for those.

/dev/shm/ is occasionally used as dedicated scratch space for some
purposes.  In fact, it's the default write location for the temporary
message in the xymon-client RPM for a while now (this way you can
actually get a report when /tmp/ fills to 100%).  With the proliferation
of spurious tmpfs mounts now, it might be a good idea to take another look.