[Bug 1987663] Re: cinder-volume: "Failed to re-export volume, setting to ERROR" with "tgtadm: failed to send request hdr to tgt daemon, Transport endpoint is not connected" on service startup

[Mauricio Faria de Oliveira]

Hi Heather,

Ok, thanks!

Let's hold on Jammy and Focal for your testing, before we upload those, per SRU policy
(template: "It is assumed that any SRU candidate patch is well-tested before upload") [1].

If you have any issues or need help, please just let me know!

[1] https://wiki.ubuntu.com/StableReleaseUpdates

-- 
You received this bug notification because you are a member of Ubuntu
OpenStack, which is subscribed to cinder in Ubuntu.
https://bugs.launchpad.net/bugs/1987663

Title:
  cinder-volume: "Failed to re-export volume, setting to ERROR" with
  "tgtadm: failed to send request hdr to tgt daemon, Transport endpoint
  is not connected" on service startup

Status in Ubuntu Cloud Archive:
  Fix Released
Status in Ubuntu Cloud Archive antelope series:
  New
Status in Ubuntu Cloud Archive bobcat series:
  New
Status in Ubuntu Cloud Archive caracal series:
  Fix Released
Status in Ubuntu Cloud Archive ussuri series:
  New
Status in Ubuntu Cloud Archive victoria series:
  New
Status in Ubuntu Cloud Archive wallaby series:
  New
Status in Ubuntu Cloud Archive xena series:
  New
Status in Ubuntu Cloud Archive yoga series:
  New
Status in Ubuntu Cloud Archive zed series:
  New
Status in cinder package in Ubuntu:
  Fix Released
Status in cinder source package in Bionic:
  Won't Fix
Status in cinder source package in Focal:
  Incomplete
Status in cinder source package in Jammy:
  Incomplete
Status in cinder source package in Lunar:
  Won't Fix
Status in cinder source package in Mantic:
  Incomplete
Status in cinder source package in Noble:
  Fix Released

Bug description:
  [Impact]

   * The cinder-volume service might fail to re-export volumes
     in-use on startup if tgt.service isn't fully started yet.

   * This affects the 'lvm' driver with 'tgtadm' target helper
     (which runs 'tgtadm' commands that need the service ready).

   * Snippets from /var/log/cinder/cinder-volume.log:

     Failed to re-export volume, setting to ERROR.
     ...
     Command: tgtadm --lld iscsi --op show --mode target
     ...
     Stderr: 'tgtadm: failed to send request hdr to tgt daemon,
     Transport endpoint is not connected\n'

   * This issue is more common in openstack compute nodes
     with networking (ovs/ovn) that takes long to startup,
     which might delay the startup of tgt.service _after_
     cinder-volume.service.

  [Test Steps]

   * Steps to reproduce are detailed in comment #3.
     Summary:

   * Install mysql, rabbitmq-server, keystone, and cinder
     (controller and storage nodes; backup node unneeded).

   * Configure cinder-volume (storage node) for LVM backend
     and tgtadm iSCSI helper (tgt.service).

   * Create a cinder volume, and configure it as 'in-use'.

   * Simulate a start delay on tgt.service with a drop-in.

   * Restart services: cinder-volume.service tgt.service

   * Check sequence of service startup.

   * Check status of the cinder volume:
     'in-use' (expected) or 'error' (bug).

   * Check log file /var/log/cinder/cinder-volume.log for
     'tgtadm: failed to send request hdr to tgt daemon'.

  [Regression Potential]

   * The fix introduces systemd unit 'After=' and 'Wants='
     properties for tgt.service in cinder-volume.service,
     thus might delay the boot process (multi-user.target).

       $ systemctl show cinder-volume.service | grep WantedBy=
       WantedBy=multi-user.target

   * However, the boot process already waits on tgt.service
     anyway, thus the difference (if any) should not be big,
     and would provide more correct behavior.

       $ systemctl show tgt.service | grep WantedBy=
       WantedBy=multi-user.target

   * If tgt.service is not present (tgt package not installed)
     _no errors_ occur, as both 'After=' and 'Wants=' are weak
     ordering/dependency properties (man 5 systemd.unit).

  [Other Info]

   * The fix uses a systemd service drop-in snippet because
     the service unit is generated by openstack-pkg-tools
     (pkgos-gen-systemd-unit) based on the 'init' service,
     and it only emits 'Wants=' for network-online.target.

   * Changing that in openstack-pkg-tools changes behavior
     in stable releases, and only manifest at build time,
     for many openstack packages that have no issues now.

   * We'll continue to pursue the general improvement in
     Debian, so it comes into Ubuntu development release,
     but for the Ubuntu stable releases, this should do.

  [Original Bug Description]

  Real-world example in comment #2.

To manage notifications about this bug go to:
https://bugs.launchpad.net/cloud-archive/+bug/1987663/+subscriptions