Pre-upgrade check

[DaveKay]

Whilst running the upgrade from 10.4.4 #36 on Ubuntu 1.0.04LTS the installer displayed a number of warnings in a pre-upgrade check. One of these was for PHP deprecation issues - no major problem. All upgraded fine.

There was another warning with a link to a KB article regarding sw-collectd and stats collection using a large amount of CPU time. Unfortunately, I stupidly ignored this warning and overnight my CPU usage has gone through the roof, almost stalling Plesk and the server.

I can confirm that the culprit processes are sw-collectd and java - they seem to have really high CPU% in TOP.

Can anyone point me in the direction of the KB article as I can't seem to find it when searching.

Any other advice is appreciated!

Many thanks,
Dave

 

[Nikolay.]

Take a look at http://forum.parallels.com/showthread.php?p=629338

Seems like you have a leap second problem with Java. Regarding sw-collectd - either you didn't notice high CPU usage before, or it's like a placebo effect (no offense ).

 

[DaveKay]

Hi Nikolay,

Server reboot seemed to sort out the Java high CPU issue. Thanks.

The sw-collectd issue has also gone for the mean time - this has never previously used more than a few % in the past. I'll wait and see if it spikes again over night! Strangely though, Health Monitor is now showing all 0 stats! Working on a fix as I type!

Cheers

 

[Faris Raouf]

http://kb.parallels.com/en/113711 is the KB article regarding sw-collectd

Essentially you just need to edit /etc/sw-collectd/collectd.conf and add "Interval 120" so it looks like this at the top:

# Parallels Panel Config file for sw-collectd.
#
# ATTENTION!!! Autogenerated file!!!

BaseDir "/usr/local/psa/var/health/data"
#TypesDB "/etc/sw-collectd/types.db"
FQDNLookup true
Hostname localhost
Interval 120

#PIDFILE    /var/run/collectd.pid
#PluginDir  /usr/lib/collectd

This is assuming you don't already have an Interval line, of course.

 

[DaveKay]

Hi Faris,

I've currently got it set at 300 Seconds - which is fine for my needs.

Many thanks!

 

[Faris Raouf]

I've also just noticed that Nikolay has posted some extremely useful info on this topic here http://forum.parallels.com/showthread.php?t=260744

 

[Faris Raouf]

Just want to underline that Nikolay's post related to fixing the 0 stats in the health monitor*. Really userful info.

*err..mostly. I think. I'm still testing. I'm getting some figures, but not others.

 

[DaveKay]

Yes, really helpful info!

After much config and reboots, I finally got Health Monitor to display some stats!

Slightly confusing when I got the following! http://www.dtphosting.co.uk/stats.jpg

But after about 5 mins, they levelled out at where they should be!

 

[Faris Raouf]

Hehe. I had some strange results at first too.

Unfortunately I'm not gettting disk stats at all.

And memory stats are a bit odd. Some of the charts have some stats for memory, but 0 is shown in the summary stats. I'll leave it overnight and see what happens.

I've tried it on four systems with similar results. Most odd.

At least it has reduced the load on the systems

 

[Nikolay.]

DaveKay, I hope you've dropped your previous statistics just to be safe. I observed a case when (without doing that) Health monitor showed enormous values at first, then normal and then was back to showing zeros. I can't be 100% sure this was because previous statistics haven't been dropped, but that usually seems to help.

BTW, did you have large numbers only for CPU usage? How many CPU cores do you have?

Faris Raouf, in the first 5 to 20 minutes statistics can indeed be odd. Please tell us if the strange behavior persists.

 

[Faris Raouf]

Well, I just can't seem to get memory stats to display. This is after many hours.

The stats are collected -- they are visible in the Detail charts. But they don't appear as summary figures.

The screen shots attached illustrate this for the memory stats under Services.

Same problem goes for Disk, total Memory and, I think, some CPU processes.

I've repeated the "stop - delete - restart" process and restarted the system to no avail.

But if I comment out the Interval line, all the stats appear again as before (with corresponding annoyingly high load)

All very interesting! I just don't get it.

Please can I emphasise that this is with Plesk 10 in my case. This is the Plesk 11 section of the forum, and the topic was about the pre-upgrade check script results. It is not my intention to confuse anyone or to take us massively off-topic. I'm just glad things are working for DaveKay which was the purpose of my original reply.

 

[DaveKay]

Nikolay, yes I removed any previously collected stats whilst sorting out the issue.

All seems fine now, stats have been stable since last night.

Yes, it was only the CPU stats that were high. This I think was partly due to the Perl leap second issue a few days ago - a reboot sorted that out. And the stats interval issue.

Just as a note, I hadn't rebooted the server after upgrading to Plesk 11. Maybe the upgrade had "confused" the stats collection process?

Following various parts of http://forum.parallels.com/showthread.php?t=260744 seemed to help.

Faris, hope you get a fix soon!

 

[Nikolay.]

DaveKay, I'm glad you've sorted it out! As for upgrade affecting statistics collection process - I don't think it did. At least not by itself. It could've affected the GUI side, however.

Faris Raouf, after some experimentation I think I figured it out. Usage values for some resource types in Plesk 10 are queried for last 5 minutes minute or so. In Plesk 11 this interval is definitely bigger. sw-collectd will attempt to put a value for each resource type in DB (.rrd) once in an Interval. Unless the value have already been written, it is read as NaN which, I suppose, gets converted to 0 for an end user of Health monitor. Another peculiarity is that several subsequent values may be NaN as sw-collectd doesn't actually guarantee providing a value once in an Interval.

Now, assuming that you've set your Interval to 300 seconds and Plesk queries usage values only for last 300 seconds, it will almost always show you zeros in summary figures. To fix this you will need to lower your Interval value to 150 30 or less so you'll have at least two intervals fit into query period. This way you'll increase a chance of getting anything useful by Plesk from sw-collectd.

Interestingly, in KB for Plesk 10 they currently advice to set Interval to 120: http://kb.parallels.com/en/114138

Hope this helps.

[Edit] I revisited the results and it seems the lowest query period is 1 minute. I've updated the post to reflect it. This is quite sad. You can try setting your Interval to 30 in hope that it will be large enough to drop CPU consumption. Please tell us if it will help you.

Sorry for using color instead of strike-out to mark changed places in the post.

 

[Nikolay.]

BTW, it is possible to artificially increase a query period used by Plesk by hacking together a simple rrdtool wrapper that will filter its arguments. Tell me if you're interested and need details.