Comment 0 for bug 1730717

This is impacting us for ubuntu autopkgtests. Eventually the whole region ends up dying because each worker is hit by this bug in turn and backs off until the next reset (6 hourly).

Guests are sometimes failing to reboot. When this happens, you see the following in the console

  [[0;32m OK [0m] Reached target Shutdown.
  [ 191.698969] watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [systemd:1]
  [ 219.698438] watchdog: BUG: soft lockup - CPU#0 stuck for 22s! [systemd:1]
  [ 226.702150] INFO: rcu_sched detected stalls on CPUs/tasks:
  [ 226.704958] »(detected by 0, t=15002 jiffies, g=5347, c=5346, q=187)
  [ 226.706093] All QSes seen, last rcu_sched kthread activity 15002 (4294949060-4294934058), jiffies_till_next_fqs=1, root ->qsmask 0x0
  [ 226.708202] rcu_sched kthread starved for 15002 jiffies! g5347 c5346 f0x2 RCU_GP_WAIT_FQS(3) ->state=0x0

One host that exhibits this behaviour was:

  Linux klock 4.4.0-98-generic #121-Ubuntu SMP Tue Oct 10 14:24:03 UTC 2017 x86_64 x86_64 x86_64 GNU/Linux

guest running:

  Linux version 4.13.0-16-generic (buildd@lcy01-02) (gcc version 7.2.0 (Ubuntu 7.2.0-8ubuntu2)) #19-Ubuntu SMP Wed Oct 11 18:35:14 UTC 2017 (Ubuntu 4.13.0-16.19-generic 4.13.4)

The affected cloud region is running the xenial/Ocata cloud archive, so the version of qemu-kvm in there may also be relevant.

Here's how I reproduced it in lcy01:

  $ for n in {1..30}; do nova boot --flavor m1.small --image ubuntu/ubuntu-artful-17.10-amd64-server-20171026.1-disk1.img --key-name testbed-`hostname` --nic net-name=net_ues_proposed_migration laney-test${n}; done
  $ <ssh to each instance> sudo reboot
  # wait a minute or so for the instances to all reboot
  $ for n in {1..30}; do echo "=== ${n} ==="; nova console-log laney-test${n} | tail; done

On bad instances you'll see the "soft lockup" message - on good it'll reboot as normal.

We've seen good and bad instances on multiple compute hosts - it doesn't feel to me like a host problem but rather a race condition somewhere that's somehow either triggered or triggered much more often by what lcy01 is running. I always saw this on the first reboot - never on first boot, and never on n>1th boot. (But if it's a race then that might not mean much.)

I'll attach a bad and a good console-log for reference.

If you're at Canonical then see internal rt #107135 for some other details.