debian: use tmpfs on /tmp + bump /tmp size on fedora #351
Conversation
There was a problem hiding this comment.
Suggestions inline, but not a real review because I'm not 100% sure how/if these will work (e.g., if debian already has a /tmp fstab entry, and other logistics).
Finally: the process for working in this repo is:
$ make IMG_SFX
$ git add IMG_SFX <------- you have to commit it every single time, on every pushThis is so incredibly annoying, that I encourage you to save this into .git/hooks/pre-push:
#!/bin/bash
#
# 2024-01-25 esm
#
# FIXME: figure out if this is a PR push or a tag push
#
imgsfx=$(<IMG_SFX)
imgsfx_history=.imgsfx.history
if [[ -e $imgsfx_history ]]; then
if grep -q "$imgsfx" $imgsfx_history; then
echo "FATAL: $imgsfx has already been used" >&2
echo "Please rerun 'make IMG_SFX'" >&2
exit 1
fi
fi
echo $imgsfx >>$imgsfx_history
cache_images/debian_setup.sh
Outdated
| # they don't have to write to persistent disk. | ||
| # https://github.com/containers/podman/pull/22533 | ||
| # We could create a systemd mount unit for it but appending fstab is easier. | ||
| $SUDO sh -c 'echo "tmpfs /tmp tmpfs size=75%,mode=1777 0 0" >> /etc/fstab' |
There was a problem hiding this comment.
I think the standard way to do this is via $SUDO tee -a:
echo "tmpfs ...." | $SUDO tee -a /etc/fstab
cache_images/fedora_setup.sh
Outdated
| # Make /tmp tmpfs bigger, by default we only get 50%. Bump it to 75% so the tests have more storage. | ||
| # Do not use 100% so we do not run out of memory for the process itself if tests start leaking big | ||
| # files on /tmp. | ||
| $SUDO mkdir /etc/systemd/system/tmp.mount.d |
There was a problem hiding this comment.
Suggestion: mkdir -p, so command won't fail if (some day) that directory already exists
cache_images/fedora_setup.sh
Outdated
| # Do not use 100% so we do not run out of memory for the process itself if tests start leaking big | ||
| # files on /tmp. | ||
| $SUDO mkdir /etc/systemd/system/tmp.mount.d | ||
| $SUDO sh -c 'echo -e "[Mount]\nOptions=size=75%%,mode=1777\n" > /etc/systemd/system/tmp.mount.d/override.conf' |
There was a problem hiding this comment.
Same tee suggestion as above.
|
Also, did you look into bumping |
well yeah that is what I figured but it gave a bunch of timebomb warning that I should fix so I didn't bother and called it a day..., will continue on Thursday |
|
Oh yeah. Easiest solution is to just bump the timebombs up by a month. |
That is the next step, but I think 4 Gb RAM is enough (if we remove the gigantic toolbox image). |
I could make it a systemd unit like on fedora then we know it will always overwrite it (Note that on systemd systems it will parse the fstab file and convert to mounts) I am not sur ehow it interacts with the systemd parser but traditionally it would just mount line by line, so even if there would be a tmpfs mount specified before it would have just over mounted it again AFAIK. |
|
Cirrus CI build successful. Found built image names and IDs:
|
|
To make tests faster setup a tmpfs on /tmp like fedora does so that test do not have to write everything onto persistent disk. Fixes containers#350 Signed-off-by: Paul Holzinger <[email protected]>
By default we only get 50% of all memory, given our programs don't take this much we should instead use more /tmp space in case we have to store more images. Signed-off-by: Paul Holzinger <[email protected]>
|
Cirrus CI build successful. Found built image names and IDs:
|
|
Not sure what to use as baseline, so I choose the last result of #349. Only one diff:
|
Luap99
force-pushed
the
debian-tmpfs
branch
2 times, most recently
from
264ba1d to
a531bef
Compare
|
Cirrus CI build successful. Found built image names and IDs:
|
|
Comparing against the previous build in this PR:
|
Now that we use /tmp we do not have to include the changes for /var/tmp. However we need r (read) access to /tmp as pasta opens the path with read access. Signed-off-by: Paul Holzinger <[email protected]>
Signed-off-by: Paul Holzinger <[email protected]>
|
Cirrus CI build successful. Found built image names and IDs:
|
|
Again, comparing against the previous build in this same PR. New pasta in f40 and rawhide, be careful
|
from containers/automation_images#351 Signed-off-by: Paul Holzinger <[email protected]>
|
@edsantiago PTAL and merge, images were merged in podman containers/podman#22533 |
|
/lgtm @Luap99 do you know the debian pasta status? Should I tag this (so renovatebot pulls it into every repo), or should we wait? |
If it passes podman I see no reason why not. AFAIK pasta is only used in buildah and podman so I don't see any particular risk with pasta on debian. |
To make tests faster setup a tmpfs on /tmp like fedora does so that test
do not have to write everything onto persistent disk.
Fixes #350