Building ISO in RAM often fails: mksquashfs reaped by OOM killer

My local builds with TAILS_BUILD_OPTIONS=ram, that use gzip SquashFS compression by default, often fail like this:

kernel: Out of memory: Kill process 12575 (mksquashfs) score 145 or sacrifice child
kernel: Killed process 12575 (mksquashfs) total-vm:3580684kB, anon-rss:1992044kB, file-rss:4kB, shmem-rss:2288kB
kernel: oom_reaper: reaped process 12575 (mksquashfs), now anon-rss:0kB, file-rss:0kB, shmem-rss:0kB

It seems that xz SquashFS compression is not affected because this does not happen on Jenkins.

When building in RAM we’re mounting a tmpfs with size=100%, which can potentially trigger this kind of issues if that filesystem is almost filled. Bumping the amount of RAM given to the VM (via VM_MEMORY_BASE) should fix it; this has been done on feature/buster already and our CI was adjusted for it already. Now, what’s strange is the fact this problem only happens ~20-40% of the time. So perhaps a system event eats precious RAM at the exact wrong time and we could fix the problem by disabling it or ensuring it’s done before the build starts.

Feature Branch: bugfix/16177-limit-mksquashfs-memory-usage

Related issues

• Related to #9788 (closed)
• Related to #16457 (closed)
• Blocks #15507 (closed)

Original created by @intrigeri on 16177 (Redmine)