yahoo-eng-team team mailing list archive

[Rich <1948459@xxxxxxxxxxxxxxxxxx>]

Public bug reported:

ZTS does a lot of spawning of relatively short-lived zvols, and the
cloud-init hotplug udev rule seems to be causing them to stay open for
much longer than the test suite needs them to be, leading to a whole
bunch of failures when the scripts expect the devices to actually go
away after being destroyed.

Inserting a blanket 'KERNEL=="zd*", GOTO="cloudinit_end"' before
cloudinit_hook seems to avoid that, but not understanding well what the
hotplug functionality wants out of life, this might be undesirable for
general use.

(If you'd like to see an example, go look at every zfs-
tests-{functional,sanity} GH Action run on OpenZFS for the past few
weeks since the change rolled out, or just run it yourself with, say,
zfs-tests.sh -T history,reservation,snapused,zfs_destroy,zfs_load-
key,zfs_receive,zfs_unmount,zfs_upgrade .)

Busywait workarounds are being added to make the tests just retry on
encountering things still being open, but "holds open disk devices for a
little while" seems like an unfortunate side effect that might surprise
some people beyond just running this test suite.

** Affects: cloud-init
     Importance: Undecided
         Status: New

-- 
You received this bug notification because you are a member of Yahoo!
Engineering Team, which is subscribed to cloud-init.
https://bugs.launchpad.net/bugs/1948459

Title:
  cloud-init hotplug hook breaks ZFS Test Suite

Status in cloud-init:
  New

Bug description:
  ZTS does a lot of spawning of relatively short-lived zvols, and the
  cloud-init hotplug udev rule seems to be causing them to stay open for
  much longer than the test suite needs them to be, leading to a whole
  bunch of failures when the scripts expect the devices to actually go
  away after being destroyed.

  Inserting a blanket 'KERNEL=="zd*", GOTO="cloudinit_end"' before
  cloudinit_hook seems to avoid that, but not understanding well what
  the hotplug functionality wants out of life, this might be undesirable
  for general use.

  (If you'd like to see an example, go look at every zfs-
  tests-{functional,sanity} GH Action run on OpenZFS for the past few
  weeks since the change rolled out, or just run it yourself with, say,
  zfs-tests.sh -T history,reservation,snapused,zfs_destroy,zfs_load-
  key,zfs_receive,zfs_unmount,zfs_upgrade .)

  Busywait workarounds are being added to make the tests just retry on
  encountering things still being open, but "holds open disk devices for
  a little while" seems like an unfortunate side effect that might
  surprise some people beyond just running this test suite.

To manage notifications about this bug go to:
https://bugs.launchpad.net/cloud-init/+bug/1948459/+subscriptions