So, I have a server at home that has four VMs running inside it.
All are managed via puppet. The physical host runs puppetmasterd.
I don't recall noticing this before, but puppetmasterd has decided
to be kind of crazy. Here's the physical host with no puppetmasterd
running:
top - 11:36:15 up 271 days, 15:16, 1 user, load average: 5.68, 5.50, 6.45
Tasks: 129 total, 1 running, 128 sleeping, 0 stopped, 0 zombie
Cpu(s): 3.6%us, 1.8%sy, 0.0%ni, 80.4%id, 14.3%wa, 0.0%hi, 0.0%si, 0.0%st
Mem: 8128776k total, 6991020k used, 1137756k free, 408756k buffers
Swap: 8388604k total, 552356k used, 7836248k free, 185220k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
10296 qemu 20 0 2366m 1.5g 8884 S 6.2 19.6 6:37.65 qemu-kvm
17334 qemu 20 0 2788m 1.7g 544 S 2.7 22.3 4576:25 qemu-kvm
9904 qemu 20 0 2358m 581m 8820 S 0.9 7.3 3:55.78 qemu-kvm
1 root 20 0 46880 8076 1372 S 0.0 0.1 0:27.00 systemd
2 root 20 0 0 0 0 S 0.0 0.0 0:10.48 kthreadd
3 root 20 0 0 0 0 S 0.0 0.0 322:04.84 ksoftirqd/0
6 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration/0
7 root RT 0 0 0 0 S 0.0 0.0 0:11.57 watchdog/0
8 root RT 0 0 0 0 S 0.0 0.0 0:00.00 migration/1
10 root 20 0 0 0 0 S 0.0 0.0 551:03.31 ksoftirqd/1
And here it is with puppetmasterd running:
top - 11:25:07 up 271 days, 15:05, 1 user, load average: 12.59, 8.68, 7.05
Tasks: 131 total, 3 running, 128 sleeping, 0 stopped, 0 zombie
Cpu(s): 15.2%us, 36.4%sy, 0.0%ni, 6.6%id, 39.7%wa, 0.0%hi, 2.0%si, 0.0%st
Mem: 8128776k total, 6830276k used, 1298500k free, 381356k buffers
Swap: 8388604k total, 555328k used, 7833276k free, 180096k cached
PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
10660 puppet 20 0 214m 107m 4040 S 61.9 1.3 8:46.81 puppetmasterd
3 root 20 0 0 0 0 S 21.4 0.0 320:38.54 ksoftirqd/0
10 root 20 0 0 0 0 R 20.2 0.0 549:30.88 ksoftirqd/1
10296 qemu 20 0 2470m 1.4g 8888 S 13.1 18.1 4:23.70 qemu-kvm
17334 qemu 20 0 2788m 1.7g 540 S 8.3 22.0 4574:54 qemu-kvm
9904 qemu 20 0 2422m 572m 8820 S 3.6 7.2 3:07.15 qemu-kvm
24980 qemu 20 0 1824m 1.4g 612 S 3.6 18.3 15046:11 qemu-kvm
12209 rlpowell 20 0 15256 1228 908 R 1.2 0.0 0:00.04 top
1 root 20 0 46880 7992 1356 S 0.0 0.1 0:26.97 systemd
2 root 20 0 0 0 0 S 0.0 0.0 0:10.48 kthreadd
The high CPU use by puppetmasterd is bad enough, but what makes me
be all like "wait, what?" is the ksoftirqd usage.
Puppet master version is 2.16.
This is *without* a client running; there's no traffic on 8140
according to tcpdump, and there's nothing happening in the log.
http://users.digitalkingdom.org/~rlpowell/media/public/puppetmasterd_strace.txt
has strace output; it's pretty boring, but there are a few select
and rt_sigprocmask calls near the bottom.
I'm totally stumped here. Any ideas?
-Robin
--
http://singinst.org/ : Our last, best hope for a fantastic future.
.i ko na cpedu lo nu stidi vau loi jbopre .i danfu lu na go'i li'u .e
lu go'i li'u .i ji'a go'i lu na'e go'i li'u .e lu go'i na'i li'u .e
lu no'e go'i li'u .e lu to'e go'i li'u .e lu lo mamta be do cu sofybakni li'u