Project

General

Profile

Actions
Copy link

Bug #6614

open

Dashboard high CPU usage

Added by Harry Coin over 8 years ago. Updated over 8 years ago.

Status:
Confirmed
Priority:
Normal
Assignee:
-
Category:
Web Interface
Target version:
-
Start date:
07/14/2016
Due date:
% Done:

0%

Estimated time:
Plus Target Version:
Release Notes:
Affected Version:
2.3.x
Affected Architecture:

Description

In a very low traffic sandbox environment, the new 2.3.2 snapshot of today hums along with a load average of 0.48 or so on a dual processor xeon sandbox with nobody logged in using the gui. I load the GUI and the load average jumps to 1.4 for a bit, then zips up to 3,x to q5.1 or so and stays locked there for as long as I'm logged in. Thing is really, really slow. All I'm doing is watching the dashboard.

last pid: 24693; load averages: 3.68, 2.40, 2.79 up 0+01:06:58 12:20:14
56 processes: 5 running, 51 sleeping
CPU: 1.8% user, 0.0% nice, 91.8% system, 5.3% interrupt, 1.2% idle
Mem: 74M Active, 81M Inact, 131M Wired, 96K Cache, 332M Buf, 1674M Free
Swap: 3072M Total, 3072M Free

PID USERNAME  THR PRI NICE   SIZE    RES STATE   C   TIME    WCPU COMMAND
 9331 root        1  82    0 16868K  5204K RUN     0   0:04  34.47% pfctl
14712 root        1  77    0   266M 34604K CPU1    1   0:14  17.19% php-fpm
30447 root        1  77    0   262M 38832K RUN     1   0:09  16.55% php-fpm
44294 root        1  52    0   266M 36592K piperd  0   0:12  11.87% php-fpm
33256 root        1  52    0   262M 38572K piperd  1   0:09   4.59% php-fpm
  264 root        1  20    0   262M 24984K kqread  1   0:04   0.10% php-fpm
35943 root        1  20    0 16676K  2244K bpf     0   0:02   0.10% filterlog
 6902 root        1  52   20 17000K  2580K wait    0   1:09   0.00% sh

Sometimes pfctl jumps to 40% load for a while and the php-fpm processes get less. When pfctl drops off the list, the php processes scale up. I'm going to delete items off the dashboard, to see if one of them is a culprit.

So... first candidate to delete: Interface Statistics. Nope. Still hanging up there, with 2 pfctls now. momentarily dtops back to 28% idle.. so, not interface stats.

Load balancer stats... gone. No obvious changes. 1.5% idle.

CARP stats... gone.. No obvious changes.

OpenVPN stats... gone. No obvious changes.

Interfaces... gone. No obvious changes.

Gateways ... gone... no obvious changes.

System information ... gone... the dashboard put it back....90% idle. There's the problem.

Maybe it's the scanner for the update? I'm told now that 2.3.2.a.20160714.1031 is available

I'm thinking maybe a memory leak / wedge in the dashboard system information panel?

Actions
Copy link
 #1

Updated by Harry Coin over 8 years ago

I visit the package manager listing, loads hover around 1, processor 98% idle. Back to the dashboard: wait 20 sec for it all to settle and it's hovering now at 5.9% idle.

last pid: 38327; load averages: 2.29, 2.07, 1.47 up 0+00:35:19 13:14:29
58 processes: 6 running, 52 sleeping
CPU: 8.8% user, 0.0% nice, 84.2% system, 5.5% interrupt, 1.6% idle
Mem: 92M Active, 131M Inact, 272M Wired, 212M Buf, 1466M Free
Swap: 3072M Total, 3072M Free

Message from syslogd@gate1 at Jul 14 13:06:55 ... C TIME WCPU COMMAND
13189 root 1 52 0 262M 35616K RUN 1 0:02 10.25% php-fpm
98398 root 1 52 0 262M 35648K ppwait 1 0:02 10.25% php-fpm
78057 root 1 52 0 262M 38044K select 1 0:02 7.67% php-fpm
22394 root 1 29 0 262M 35420K accept 0 0:01 3.86% php-fpm
82631 root 1 52 20 17000K 2588K wait 1 0:39 0.20% sh

Every now and then a zombie pops up and the load decreases, then it goes back to the above:

last pid: 49149; load averages: 2.49, 2.15, 1.52 up 0+00:35:55 13:15:05
52 processes: 2 running, 49 sleeping, 1 zombie
CPU: 7.2% user, 3.1% nice, 62.5% system, 7.4% interrupt, 19.9% idle
Mem: 85M Active, 127M Inact, 270M Wired, 213M Buf, 1480M Free
Swap: 3072M Total, 3072M Free

PID USERNAME  THR PRI NICE   SIZE    RES STATE   C   TIME    WCPU COMMAND
 9541 root        1  34    0   262M 35556K nanslp  0   0:02  10.35% php-fpm
82631 root        1  93   20 17000K  2588K CPU1    1   0:39   4.88% sh
34075 root        1  20    0 21620K  5736K select  1   0:06   0.00% openvpn

last pid: 71403; load averages: 2.43, 2.25, 1.61 up 0+00:37:06 13:16:16
55 processes: 4 running, 50 sleeping, 1 zombie
CPU: 3.7% user, 0.0% nice, 87.5% system, 3.9% interrupt, 5.0% idle
Mem: 89M Active, 127M Inact, 271M Wired, 213M Buf, 1474M Free
Swap: 3072M Total, 3072M Free

PID USERNAME  THR PRI NICE   SIZE    RES STATE   C   TIME    WCPU COMMAND
35987 root        1  79    0   262M 35696K RUN     1   0:04  24.07% php-fpm
50532 root        1  52    0   262M 35600K piperd  0   0:01   8.15% php-fpm
68827 root        5  20    0 15012K  2292K accept  0   0:01   0.10% dpinger
82631 root        1  52   20 17000K  2588K wait    1   0:40   0.00% sh

Loads spike to 5 when the pfctl process gets going to deal with a net interface going offline.

Actions
Copy link
 #2

Updated by Chris Buechler over 8 years ago

  • Subject changed from Dashboard takes load av from 0.45 to 5+ to Dashboard high CPU usage
  • Status changed from New to Confirmed
  • Target version deleted (2.3.2)
  • Affected Version changed from 2.3.2 to 2.3.x
  • Affected Architecture added
  • Affected Architecture deleted (amd64)

the number of things that dynamically update is significantly higher than it was in 2.2.x and prior. Still, something someone should look at, as it seems excessive in some circumstances.

Actions
Copy link

Also available in: Atom PDF