[Check_mk (english)] probable bug in 1.5.0 perfdata if64

[Dr. Maik Götte]

Hi all

My networking colleagues are experiencing issues with the if64 perfdata.
We are monitoring our network devices once every 10 minutes. Until the
upgrade to 1.5 (from 1.4) everything went smooth. Now we observe the
following strange behaviour:

Real Perfdata seems to be stored in RRDs (and therefore probably in
perfometer also) every 10 minutes. Inbetween it seem's it is filled by
zeros, not the last measured data. This leads to "spikey" graphs and a
perfometer which only shows data every 10 minutes. This "filling" seems
to be somehow hardcoded to 1 minute. If we switch to a 1 minute
monitoring intervall everything seems fine.

The behaviour of other tests like CPU util or some such is ok, so it
seems to only affect the interfaces. This issue probably arose with the
rewrite of the respective python code. I had a quick look into it but
could'nt find anything.

Are there people out there who experience the same issues or maybe even
have a fix?

You can find screenshots for cpu and if of the same machine here:

Loading Image...
Loading Image...


Best regards
Maik

[Wilke Alexander]

Hello Maik,

did you only tune down the interval for the check to 10 Minutes, but you did not change and recreate the RRDs?
Then the data is checked every 10 Minutes, but the rrds are filled with NULL values, because the heartbeat of the rrd is set to less than 1 Minute.

Alex

-----Ursprüngliche Nachricht-----
Von: checkmk-en <checkmk-en-***@lists.mathias-kettner.de> Im Auftrag von Dr. Maik Götte
Gesendet: Donnerstag, 18. Oktober 2018 08:37
An: checkmk-***@lists.mathias-kettner.de
Betreff: [Check_mk (english)] probable bug in 1.5.0 perfdata if64

Hi all

My networking colleagues are experiencing issues with the if64 perfdata.
We are monitoring our network devices once every 10 minutes. Until the upgrade to 1.5 (from 1.4) everything went smooth. Now we observe the following strange behaviour:

Real Perfdata seems to be stored in RRDs (and therefore probably in perfometer also) every 10 minutes. Inbetween it seem's it is filled by zeros, not the last measured data. This leads to "spikey" graphs and a perfometer which only shows data every 10 minutes. This "filling" seems to be somehow hardcoded to 1 minute. If we switch to a 1 minute monitoring intervall everything seems fine.

The behaviour of other tests like CPU util or some such is ok, so it seems to only affect the interfaces. This issue probably arose with the rewrite of the respective python code. I had a quick look into it but could'nt find anything.

Are there people out there who experience the same issues or maybe even have a fix?

You can find screenshots for cpu and if of the same machine here:

https://hessenbox.uni-kassel.de/dl/fiGMCQsQMDULkEAczxKZUVes/cpu_example.PNG?inline
https://hessenbox.uni-kassel.de/dl/fi4R9M1nifpVExLw5ZK5NVnL/if_example.PNG?inline


Best regards
Maik



________________________________
IABG TELEPORT GmbH
Sitz der Gesellschaft: Ottobrunn
Registergericht: Amtsgericht München, HRB 236658
Geschäftsführung: Frank Bauner

[Dr. Maik Götte]

Hi Alexander,

thank you for your answer.
Yes, we did exactly that. I came to the same conclusion as you did.
Unfortunately, the rest of our experiences is voting against that. As I
wrote, CPU for the same device DOES insert the last value for the next
10 minutes instead of zero. And with 1.4 it was the same behaviour for
interfaces. I also think that's the correct way to do this. It's not
really comfortable to change the individual RRD sampling when changing
the sampling rate for the test itself. Would be a lot of work and error
prone.

So I still guess, it's a bug. I think I will ask our support.

Best regards
Maik

Post by Wilke Alexander
Hello Maik,
did you only tune down the interval for the check to 10 Minutes, but you did not change and recreate the RRDs?
Then the data is checked every 10 Minutes, but the rrds are filled with NULL values, because the heartbeat of the rrd is set to less than 1 Minute.
Alex
-----UrsprÃŒngliche Nachricht-----
Gesendet: Donnerstag, 18. Oktober 2018 08:37
Betreff: [Check_mk (english)] probable bug in 1.5.0 perfdata if64
Hi all
My networking colleagues are experiencing issues with the if64 perfdata.
Real Perfdata seems to be stored in RRDs (and therefore probably in perfometer also) every 10 minutes. Inbetween it seem's it is filled by zeros, not the last measured data. This leads to "spikey" graphs and a perfometer which only shows data every 10 minutes. This "filling" seems to be somehow hardcoded to 1 minute. If we switch to a 1 minute monitoring intervall everything seems fine.
The behaviour of other tests like CPU util or some such is ok, so it seems to only affect the interfaces. This issue probably arose with the rewrite of the respective python code. I had a quick look into it but could'nt find anything.
Are there people out there who experience the same issues or maybe even have a fix?
https://hessenbox.uni-kassel.de/dl/fiGMCQsQMDULkEAczxKZUVes/cpu_example.PNG?inline
https://hessenbox.uni-kassel.de/dl/fi4R9M1nifpVExLw5ZK5NVnL/if_example.PNG?inline
Best regards
Maik
________________________________
IABG TELEPORT GmbH
Sitz der Gesellschaft: Ottobrunn
Registergericht: Amtsgericht MÃŒnchen, HRB 236658
GeschÀftsfÌhrung: Frank Bauner

[Dr. Maik Götte]

Hi all

I just wanted to inform you that support acknowledged this to be a bug.
There is, however, a workaround: If an additional "Monitoring
Configuration -> Service Checks -> Normal check interval for service
checks" (cmk-agent) is set accordingly to the snmp check, the RRDs and
Perfometer are behaving normal again.

Best regards
Maik

Post by Dr. Maik Götte
Hi Alexander,
thank you for your answer.
Yes, we did exactly that. I came to the same conclusion as you did.
Unfortunately, the rest of our experiences is voting against that. As I
wrote, CPU for the same device DOES insert the last value for the next
10 minutes instead of zero. And with 1.4 it was the same behaviour for
interfaces. I also think that's the correct way to do this. It's not
really comfortable to change the individual RRD sampling when changing
the sampling rate for the test itself. Would be a lot of work and error
prone.
So I still guess, it's a bug. I think I will ask our support.
Best regards
Maik

Post by Wilke Alexander
Hello Maik,
did you only tune down the interval for the check to 10 Minutes, but
you did not change and recreate the RRDs?
Then the data is checked every 10 Minutes, but the rrds are filled
with NULL values, because the heartbeat of the rrd is set to less than
1 Minute.
Alex
-----UrsprÃŒngliche Nachricht-----
Auftrag von Dr. Maik Götte
Gesendet: Donnerstag, 18. Oktober 2018 08:37
Betreff: [Check_mk (english)] probable bug in 1.5.0 perfdata if64
Hi all
My networking colleagues are experiencing issues with the if64 perfdata.
We are monitoring our network devices once every 10 minutes. Until the
upgrade to 1.5 (from 1.4) everything went smooth. Now we observe the
Real Perfdata seems to be stored in RRDs (and therefore probably in
perfometer also) every 10 minutes. Inbetween it seem's it is filled by
zeros, not the last measured data. This leads to "spikey" graphs and a
perfometer which only shows data every 10 minutes. This "filling"
seems to be somehow hardcoded to 1 minute. If we switch to  a 1 minute
monitoring intervall everything seems fine.
The behaviour of other tests like CPU util or some such is ok, so it
seems to only affect the interfaces. This issue probably arose with
the rewrite of the respective python code. I had a quick look into it
but could'nt find anything.
Are there people out there who experience the same issues or maybe even have a fix?
https://hessenbox.uni-kassel.de/dl/fiGMCQsQMDULkEAczxKZUVes/cpu_example.PNG?inline
https://hessenbox.uni-kassel.de/dl/fi4R9M1nifpVExLw5ZK5NVnL/if_example.PNG?inline
Best regards
Maik
________________________________
IABG TELEPORT GmbH
Sitz der Gesellschaft: Ottobrunn
Registergericht: Amtsgericht MÃŒnchen, HRB 236658
GeschÀftsfÌhrung: Frank Bauner

_______________________________________________
checkmk-en mailing list
Manage your subscription or unsubscribe
http://lists.mathias-kettner.de/mailman/listinfo/checkmk-en