Very high CPU usage httpd, mysqld, FOGMulticastManager FOG trunk@5224

[Kiweegie]

@Wayne-Workman Not a problem. Installed on both boxes and currently running - I need to run myself shortly but will fire over the results before i go in about 20 mins.

cheers, Kiweegie.

[Joe Schmitt]

Some info for people experiencing high loads:

v0.10.0 of the client (currently undergoing release candidate testing) drastically cuts the amount of traffic used. In the “heaviest” cycle, the client will only make 3 requests. 1 to authenticate, 1 to get module settings, and 1 to get server settings (cycle time, client version,…). It also prevents run-away authentication, where the client had the potential to constantly spam the server’s authentication method.

Hopefully when v0.10.0 is released it should resolve any load issues. However, more than likely v0.10.0 may contain several bugs as the entire code base has changed to support any OS, and a completely new middleware API has been made to support the decrease in traffic. Since so much has changed, its almost guaranteed that I will have overlooked some minor things.

[Kiweegie]

@Wayne-Workman Hi Wayne

here’s the output of apachetop on my main machine

last hit: 15:16:16         atop runtime:  0 days, 00:21:30             15:16:17
All:        22036 reqs (  17.1/sec)       4721.2K ( 3747.7B/sec)     219.4B/req
2xx:   22036 ( 100%) 3xx:       0 ( 0.0%) 4xx:     0 ( 0.0%) 5xx:     0 ( 0.0%)
R ( 30s):     454 reqs (  15.1/sec)         90.0K ( 3072.1B/sec)     203.0B/req
2xx:     454 ( 100%) 3xx:       0 ( 0.0%) 4xx:     0 ( 0.0%) 5xx:     0 ( 0.0%)

REQS REQ/S    KB KB/S URL
 210  7.00   1.0  0.0*/fog/service/servicemodule-active.php
  53  1.77  86.9  2.9 /fog/management/other/ssl/srvpublic.crt
  43  1.43   0.5  0.0 /fog/service/Printers.php
  30  1.00   0.2  0.0 /fog/service/autologout.php
  28  0.97   0.1  0.0 /fog/service/snapins.checkin.php
  17  0.59   0.9  0.0 /fog/management/index.php
  16  0.53   0.1  0.0 /fog/service/greenfog.php
  15  0.50   0.1  0.0 /fog/service/printerlisting.php
  14  0.47   0.1  0.0 /fog/service/getversion.php
  14  0.48   0.1  0.0 /fog/service/jobs.php
  13  0.45   0.1  0.0 /fog/service/hostname.php
   1  0.04   0.0  0.0 *

and from my second machine

last hit: 15:15:47         atop runtime:  0 days, 00:21:10             15:15:48
All:         5274 reqs (   4.2/sec)       1008.8K (  813.4B/sec)     195.9B/req
2xx:    5259 (99.7%) 3xx:      15 ( 0.3%) 4xx:     0 ( 0.0%) 5xx:     0 ( 0.0%)
R ( 30s):     190 reqs (   6.3/sec)         50.5K ( 1722.4B/sec)     272.0B/req
2xx:     190 ( 100%) 3xx:       0 ( 0.0%) 4xx:     0 ( 0.0%) 5xx:     0 ( 0.0%)

REQS REQ/S    KB KB/S URL
  80  2.67   0.4  0.0*/fog/service/servicemodule-active.php
  30  1.07  49.2  1.8 /fog/management/other/ssl/srvpublic.crt
  13  0.45   0.2  0.0 /fog/service/Printers.php
   9  0.35   0.0  0.0 /fog/service/hostname.php
   9  0.33   0.1  0.0 /fog/service/getversion.php
   9  0.35   0.0  0.0 /fog/service/jobs.php
   9  0.38   0.0  0.0 /fog/service/printerlisting.php
   9  0.36   0.0  0.0 /fog/service/snapins.checkin.php
   9  0.38   0.0  0.0 /fog/service/greenfog.php
   8  0.35   0.4  0.0 /fog/management/index.php
   5  0.17   0.0  0.0 /fog/service/autologout.php

Heading off home now but will be online later if more information required.

regards Kiweegie.

[Wayne Workman]

Ok, @Developers what does servicemodule-active.php do? On Kiweegie’s servers, it’s getting 7 requests per second.

[Joe Schmitt]

For each module the client has, it does 1 request to servicemodule-active to see if its enabled. That means, every 1 cycle of the client it’s calling that file about 5-6 times. This is one of the things the v0.10.0 removes completely.

Keep in mind the new client was originally designed for a completely different form of server<-> client communication (socket connections). And inorder to be able to release it in a timely fashion, we retrofitted the legacy client’s communication method onto the new client. This is why we have so much “useless” network traffic. v0.10.0 provides a hybrid of sorts between the ideal network communication model we originally wanted, and the legacy one.

[Tom Elliott]

Just pushed a fix to hopefully make things functional again (meaning cancelling/deleting/viewing) everything

[Kiweegie]

@Tom-Elliott Hey Tom, many thanks. Both servers now have access to the reports, fog settings etc pages. CPU usage still high but the servers are usable enough albeit slow. I’ll keep an eye on Jbobs progress with the new client.

thanks again, Kiweegie.

[Kiweegie]

@Tom-Elliott Hey Tom

another anomaly turned up since most recent changes. My Helpdesk team complained that all desktops with FOG client installed were boot looping. Work around was to remove all clients from the host management section in FOG. This permitted desktops to boot and not restart - suspect this is down to the old FOG client.

Only today getting time investigate this and noticed that when adding in a brand new host it seems to accept ok but nothing shows in the GUI when clicking List all hosts. If I export the list it shows one host only (I’ve added 2) but if I try and re-add the 2nd host i get error Hostname already exists.

Something funky going on with the database?

cheers Kiweegie

[Wayne Workman]

@Kiweegie go into mysql via CLI and look at the actual data, and let us see what you see.

mysql
use fog
select * from hosts;

[Wayne Workman]

@Kiweegie Lots of improvements in the last week to the new client, can you update and see if this affects the CPU load at all?

I suspect there will be a quite dramatic decrease after moving to client 010.5

[Kiweegie]

@Wayne-Workman Hi Wayne

apologies for not getting back to you til now, other projects keep piling up… I’ve checked the DB as you suggested via the cli and although only 3 hosts are listed in the GUI the cli shows 270 hosts…

|    270 | FOG01-current   |                                              |        |         3 |            0 | 2016-05-21 11:28:20 | 2016-05-21 11:44:32 | foguser        | 0         |                    |                                                    |                   |                                                                                                    |                  |                                                                                                    |                  |                |            |            |             |            |              | 0000-00-00 00:00:00 | 6            |              |             | 1           |

Above is last snippet.

I guess I can just drop the DB and start over?

cheers Kiweegie.

[Sebastian Roth]

@Kiweegie There have been a lot of improvements in the last two weeks! Mainly in the FOG client. Please upgrade to the latest version (first try the client, then FOG server) and see if things have changed for you as well.

[Kiweegie]

@Sebastian-Roth Just upgraded trunk (5554) using git now and getting this error on install

./bin/installfog.sh: line 452: installPackages: command not found

 * Confirming package installation

./bin/installfog.sh: line 456: confirmPackageInstallation: command not found

 * Configuring services


 * What is the storage location for your images directory? (/images)

[Quazz]

@Kiweegie You have to run installfog.sh while inside the bin directory otherwise it will fail in more way than one (just checked this on my installation to be sure)

[Kiweegie]

@Quazz Thanks Quazz but I am inside the bin directory as my screenshot shows I believe. It would fail well before there if not.

[Quazz]

@Kiweegie No, you ran this command

./bin/installfog.sh

which indicates you are not in the bin directory.

[Kiweegie]

@Quazz You’re quite right, apologies Quazz I stand corrected. I was in /opt/git/trunk and presumed running ./bin/installfog.sh would suffice. Must not be enough caffeine in the system this morning…

[Quazz]

@Kiweegie Any update on this? Is it better now?

[Kiweegie]

@Quazz Hi Quazz yes main problem of high cpu usage appears to be much better - main server with most number of hosts is sitting at 5-30% CPU load roughly with no active tasks running. I’m testing now with images running to double check under load.

cheers Kiweegie.