/run/netns doesn't exist until we create a network namespace

Bug #1799638 reported by Cédric Jeanneret on 2018-10-24
This bug affects 1 person
Affects Status Importance Assigned to Milestone
Cédric Jeanneret

Bug Description


Apparently the /run/netns directory doesn't exist until a network namespace is created. This prevent podman containers to start properly after a system reboot.

For the records, podman doesn't create the host directory tree when we mount a volume in a container. Hence, as /run/netns doesn't exist, podman crashes when it tries to start containers having that location bind-mounted.

A possible way to mitigate it would be to add a "ExecPreStart=ip netns add placeholder" or something like that in the systemd unit for container having that mount.

Or create a dedicated unit for that, and make the container depend on that unit.

Cédric Jeanneret (cjeanner) wrote :
Changed in tripleo:
assignee: nobody → Cédric Jeanneret (cjeanner)
status: Triaged → In Progress
Cédric Jeanneret (cjeanner) wrote :

Fix merged.

Changed in tripleo:
status: In Progress → Fix Committed
Changed in tripleo:
status: Fix Committed → Fix Released
Rabi Mishra (rabi) wrote :

This isn't fixed yet:/

Please check the errors[1] in https://review.openstack.org/#/c/630631/. It was working earlier as we were creating containers that were using podman networking (cni0). Once I changed them to use net: none, the containers that mount /run/netns started to fail.

[1] http://logs.openstack.org/31/630631/4/check/tripleo-ci-centos-7-undercloud-containers/73d9cbf/logs/undercloud/home/zuul/undercloud_install.log.txt.gz#_2019-01-14_15_05_47

Reviewed: https://review.openstack.org/630936
Committed: https://git.openstack.org/cgit/openstack/tripleo-heat-templates/commit/?id=9b284e74cbdcf43253edaf424534b71a483fdb69
Submitter: Zuul
Branch: master

commit 9b284e74cbdcf43253edaf424534b71a483fdb69
Author: Rabi Mishra <email address hidden>
Date: Tue Jan 15 16:13:33 2019 +0530

    Create /run/netns if does not exist

    Unlike docker with podman you can't mount a host directory
    that does not exist. We did some work as part of bug 1799638
    take care of this for reboots. However, we need it to be created
    for fresh installations, as nodes are not necessarily rebooted
    after installing paunch. The real issue without reboots did not
    surface before the next patch in this series, as we were using
    cni0 bridge network for some containers and that resulted in
    namespace being created and the directory existing for
    neutron_dhcp and neutron_l3 containers in step 4 to mount them.

    This patch creates/removes a temp namespace in host_prep_tasks
    to ensure that the directory exists.

    Change-Id: I19d660168c98887a5e352b3413235888c800760d
    Related-Bug: #1799638

To post a comment you must log in.
This report contains Public information  Edit
Everyone can see this information.

Other bug subscribers