the exit file
If the container exit code needs to be retained, it cannot be retained
in tmpfs, because libpod runs in a memcg itself so it can't leave
traces with a daemon-less design.
This wasn't a memleak detectable by kmemleak for example. The kernel
never lost track of the memory and there was no erroneous refcounting
either. The reference count dependencies however are not easy to track
because when a refcount is increased, there's no way to tell who's
still holding the reference. In this case it was a single page of
tmpfs pagecache holding a refcount that kept pinned a whole hierarchy
of dying memcg, slab kmem, cgropups, unrechable kernfs nodes and the
respective dentries and inodes. Such a problem wouldn't happen if the
exit file was stored in a regular filesystem because the pagecache
could be reclaimed in such case under memory pressure. The tmpfs page
can be swapped out, but that's not enough to release the memcg with
CONFIG_MEMCG_SWAP_ENABLED=y.
No amount of more aggressive kernel slab shrinking could have solved
this. Not even assigning slab kmem of dying cgroups to alive cgroup
would fully solve this. The only way to free the memory of a dying
cgroup when a struct page still references it, would be to loop over
all "struct page" in the kernel to find which one is associated with
the dying cgroup which is a O(N) operation (where N is the number of
pages and can reach billions). Linking all the tmpfs pages to the
memcg would cost less during memcg offlining, but it would waste lots
of memory and CPU globally. So this can't be optimized in the kernel.
A cronjob running this command can act as workaround and will allow
all slab cache to be released, not just the single tmpfs pages.
rm -f /run/libpod/exits/*
This patch solved the memleak with a reproducer, booting with
cgroup.memory=nokmem and with selinux disabled. The reason memcg kmem
and selinux were disabled for testing of this fix, is because kmem
greatly decreases the kernel effectiveness in reusing partial slab
objects. cgroup.memory=nokmem is strongly recommended at least for
workstation usage. selinux needs to be further analyzed because it
causes further slab allocations.
The upstream podman commit used for testing is
1fe2965e4f (v1.4.4).
The upstream kernel commit used for testing is
f16fea666898dbdd7812ce94068c76da3e3fcf1e (v5.2-rc6).
Reported-by: Michele Baldessari <michele@redhat.com>
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
<Applied with small tweaks to comments>
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
The release-task ***must*** always execute last, in order to guarantee a
consistent cache of release archives from dependent tasks. It
accomplishes this by verifying it's task-number matches one-less than
the total number of tasks. Previous to this commit, a YAML anchor/alias
was used to avoid duplication of the dependency list between 'success'
and 'release'
However, it's been observed that this opens the possibility for
'release' and 'success' tasks to race when running on a PR. Because
YAML anchor/aliases cannot be used to modify lists, duplication is
required to make 'release' actually depend upon 'success'.
This duplication will introduce an additional maintenance burden.
Though when adding a new task, it's already very easy to forget to
update the 'depends_on' list. Assist both cases by the addition
unit-tests to verify ``.cirrus.yml`` dependency contents and structure.
Signed-off-by: Chris Evich <cevich@redhat.com>
This task depends upon other tasks caching their binaries. If for
whatever reason the `release` task is re-run and/or is out-of-order
with it's dependents, the state of cache will be undefined. Previously
this would result in an error, and failing of the release task.
This commit alters this behavior to issue a warning instead.
Signed-off-by: Chris Evich <cevich@redhat.com>
close#3648
podman create and podman run do not set --env variable if the environment is not present with a value
Signed-off-by: Qi Wang <qiwan@redhat.com>
Add an error when there are not enough input arguments for remote
create. Addresses comments in #3656
Signed-off-by: TomSweeneyRedHat <tsweeney@redhat.com>
NixOS links the current system state to `/run/current-system`, so we
have to add these paths to the configuration files as well to work out
of the box.
Signed-off-by: Sascha Grunert <sgrunert@suse.com>
The regular expression used in the `info` test does not allow for
usernames that have a dash, such as `test-user`. This patch adjusts
the regex to allow for a dash.
Fixes#3666.
Signed-off-by: Major Hayden <major@redhat.com>
The function to generate random IP addresses during ginkgo tests in
the checkpoint test code is moved to common and all tests using
hardcoded IP addresses have been changed to use random IP addresses to
reduce test errors when running the tests in parallel.
Signed-off-by: Adrian Reber <areber@redhat.com>
Update the links for the asciinema casts and the demo for the
`podman pause` and `podman unpause` commands on the commands.md
page.
Signed-off-by: TomSweeneyRedHat <tsweeney@redhat.com>
It's been observed on several occasions, some tests fail in git clones
with a "cannot find ref" type error. Especially in the depth=1 cases.
Since there's really only one place where limiting the depth makes sense
(build-each-commit), simply remove all the other limits.
Signed-off-by: Chris Evich <cevich@redhat.com>
Created shell script to automatically compile remote-only docs & rename
Added make brew-pkg to automatically package files needed for homebrew
Add missing docs
Signed-off-by: Ashley Cui <ashleycui16@gmail.com>
It seems like our VM images now support systemd CGroups with the
Ubuntu LTS images. No reason to keep testing CGroupfs as such,
systemd is much less racy (and CGroupfs on systemd-enabled
systems can be iffy).
Signed-off-by: Matthew Heon <mheon@redhat.com>
It's somewhat hard to predict which units are certinly present
for any given base-image. Therefore, at image-build time, it's
distracting and unhelpful to see all the errors about units that
don't exist, on every platform. Simply ignore them and rely on
the `check_image.sh` test to confirm none are enabled.
Signed-off-by: Chris Evich <cevich@redhat.com>
...to work for specific edge cases with a simpler solution.
Re-reads hooks directories after any changes are detected by the watchers.
Added monitoring test for adding a different invalid hook to primary directory.
Some issues with prior code:
- ReadDir would stop when it encounters an invalid hook, rather than registering an error but continuing to read the valid hook.
- Wouldn’t account for Rename and Chmod events.
- After doing a mv of the hooks file instead of rm, it would still think the hooks file is in the directory, but it has been moved to another location.
- If a hook file was renamed, it would register the renamed file as a separate hook and not delete the original, so it would then execute the hook twice - once for the renamed file, and once for the original name which it did not delete.
Signed-off-by: samc24 <sam.chaturvedi24@gmail.com>
We should not be fuzzy matching on volume names. Docker doesn't
do it, and it doesn't make much sense. Everything requires exact
matches for names - only IDs allow partial matches.
Fixes#3635
Signed-off-by: Matthew Heon <matthew.heon@pm.me>
Matthew Heon
OpenShift Merge Robot
Chris Evich
Pete Johanson
Qi Wang
TomSweeneyRedHat
Sascha Grunert
Major Hayden
Laszlo Gombos
Valentin Rothberg
Adrian Reber
Jhon Honce
Giuseppe Scrivano
Ashley Cui
Matthew Heon
samc24