Closed shahzebsiddiqui closed 6 years ago
@shahzebsiddiqui I will try a fix in sregistry to the add/move
branch that you are reviewing.
okay all fixed! Please test. https://github.com/singularityhub/sregistry-cli/pull/67
Hi @shahzebsiddiqui - did the fix .@vsoch mentioned fix this? It looks like that was related to the sregistry CLI rather than Singularity itself. If you think the space-filling is due to singularity itself please let us know. Thanks!
@dctrud @vsoch So we are still running version 2.4-dist and I was building squashfs images which according to singularity writes some files in /tmp.
According to documentation http://singularity.lbl.gov/build-environment
Singularity also uses some temporary directories to build the squashfs filesystem, so this temp space needs to be large enough to hold the entire resulting Singularity image. By default this happens in /tmp but can be overridden by setting SINGULARITY_TMPDIR to the full path where you want the squashfs temp files to be stored. Since images are typically built as root, be sure to set this variable in root’s environment.
Sorry I dont think this is an issue related with sregistry but rather how singularity works. Not sure if this behavior is still present in latest release but we ran out of space in /tmp because we only have few GBs allocated for this.
I suppose I can set SINGULARITY_TMPDIR system-wide to something like lustre to avoid singularity build filling up compute nodes /tmp.
Granted that right now root is limited to HPC users, I will be building bunch of containers and I don't want my container builds to fail due to space. I think it is worth. I think we can close this issue but worth mentioning to singularity team.
@shahzebsiddiqui - thank you for your reply. Closing as you suggest, since SINGULARITY_TMPDIR will fix for you. /tmp is still used in the current release for the build process.
Version of Singularity:
Recently running into an issue where /tmp gets 100% full because of doing singularity build or pulls. not sure what is causing this. I am doing periodic builds and pulls and storing the images somewhere else but /tmp is being written. If you can identify how this happening that would be great.