| FazBrowse GitHub Viewer | Trending | | Home |
| Tools: [Download Repo ZIP] [Original HTTPS Page] |
Sorry, something went wrong.
cmd(check=True) raised CalledProcessError with none of the command's output, which made failures hard to diagnose. Log stdout and stderr on failure, and attach them to the exception. Signed-off-by: Zack Cerza <zack@cerza.org>
podman machine ssh joins its arguments into one remote shell line, so an argument like a bash -c script was split into words. Quote each one. Signed-off-by: Zack Cerza <zack@cerza.org>
doctor only checked that a ceph_devstack module was loaded, so an outdated one went unnoticed and could block containers. Compare the installed module's hash with the shipped ceph_devstack.pp, falling back to the presence check if the installed one can't be extracted. Signed-off-by: Zack Cerza <zack@cerza.org>
Every nested container gets a session keyring charged to the rootless user. The default quota (200 keys, 20000 bytes) ran out with four concurrent orch:cephadm:smoke-small jobs, failing cephadm commands at random with "crun: join keyctl ...: Disk quota exceeded". Require kernel.keys.maxkeys >= 20000 and kernel.keys.maxbytes >= 2000000. Signed-off-by: Zack Cerza <zack@cerza.org>
cephadm reads device properties from the udev database. Mount it read-only, and allow that in the SELinux policy. Signed-off-by: Zack Cerza <zack@cerza.org>
With podman 5.8 and crun 1.29 on the host, containers that cephadm started inside a testnode failed with "crun: mount `tmpfs` to `proc/acpi`: Permission denied". Unmasking /proc in the testnode lets nested containers mask it themselves. Signed-off-by: Zack Cerza <zack@cerza.org>
The teuthology container keeps the environment it was created with, so a later start ignored a newly exported TEUTHOLOGY_SUITE, and failed outright if the SSH agent socket it mounted was gone (as when create and start run in different SSH sessions): crun: cannot stat `/tmp/ssh-XXXX.../agent.NNN`: No such file or directory A stopped teuthology container holds nothing the archive doesn't, so replace it unless it is running. Also check for the agent socket on the container host rather than the local machine. Signed-off-by: Zack Cerza <zack@cerza.org>
Teuthology is expected to exit when its run completes, so watch mode should not restart it. Signed-off-by: Zack Cerza <zack@cerza.org>
| Back | FazBrowse Home | New Git URL |
Small fixes and improvements that stand on their own that the rest of the configurable stacks PRs build on.
The README fix is needed because main fails pre-commit without it. The last four came out of running orch:cephadm:smoke-small on a CentOS 9 host with podman 5.8.5 and crun 1.29.1: nested containers need /proc unmasked, four concurrent jobs exhaust the default keyring quota, and a stopped teuthology container kept a stale SSH agent socket and TEUTHOLOGY_* settings.
Each commit applies to main on its own and passes unit tests and pre-commit run --all-files.
Testing
orch:cephadm:smoke-small 4/4 passed on folio12 with these changes and the stock main build (d42b952); without them, all four jobs failed.