wasteful re-compression

[gaponenko]

Hello,

I have just waited for several minutes as jobsub_submit was
re-compressing an already compressed code tarball. It was specified
with the

--tar_file_name dropbox:///pnfs/mu2e/resilient/users/gandr/gridexport/tmp.9I7Gv1adwT/Code.tar.bz

option, and then I saw a large file named Code.tar.bz2473.tbz2 appear
in my working directory as I was waiting for the submission to
complete.

Maybe the compression step should be delegated to the user, and
jobsub_submit should not try to re-pack the user-provided file. Just
upload it as is from its original location.

Andrei

[marcmengel]

What jobsub_lite is doing is rewriting the tarfile with the permissions modified, to prevent people putting things into cvmfs that they cannot read.
See: https://github.com/fermitools/jobsub_lite/blob/master/lib/tarfiles.py#L83
The generated tarfile is compressed just to minimize the disk required.

[gaponenko]

Can it check that the provided file has proper permissions and complain and stop if not? Let the user fix their problems instead of trying to do this for them, as it penalizes other users. Andrei

…

________________________________________ From: Marc Mengel ***@***.***> Sent: Wednesday, February 28, 2024 5:30 PM To: fermitools/jobsub_lite Cc: Andrei Gaponenko; Author Subject: Re: [fermitools/jobsub_lite] wasteful re-compression (Issue #543) [EXTERNAL] – This message is from an external sender What jobsub_lite is doing is rewriting the tarfile with the permissions modified, to prevent people putting things into cvmfs that they cannot read. See: https://github.com/fermitools/jobsub_lite/blob/master/lib/tarfiles.py#L83<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_fermitools_jobsub-5Flite_blob_master_lib_tarfiles.py-23L83&d=DwMFaQ&c=gRgGjJ3BkIsb5y6s49QqsA&r=O47fc5vzDTR2V_gla4Ub0Q&m=eNHRb5RZB0WGq-pxz0CUxaQqUR4TfYC28VJFDwjFQNp9QFyKJw7FSMxMh83va0ag&s=78usikqAjfXIDn0pS29hd6u3SqV-9taGJQE_evqitlI&e=> The generated tarfile is compressed just to minimize the disk required. — Reply to this email directly, view it on GitHub<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_fermitools_jobsub-5Flite_issues_543-23issuecomment-2D1970089983&d=DwMFaQ&c=gRgGjJ3BkIsb5y6s49QqsA&r=O47fc5vzDTR2V_gla4Ub0Q&m=eNHRb5RZB0WGq-pxz0CUxaQqUR4TfYC28VJFDwjFQNp9QFyKJw7FSMxMh83va0ag&s=CTPr_PlME8hwE1HGbGk2HlUf36P0vRLjscUcwAfGIxA&e=>, or unsubscribe<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AAXVCGU7CJASTGNJBZCUTCDYV64XZAVCNFSM6AAAAABD64YGM2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZQGA4DSOJYGM&d=DwMFaQ&c=gRgGjJ3BkIsb5y6s49QqsA&r=O47fc5vzDTR2V_gla4Ub0Q&m=eNHRb5RZB0WGq-pxz0CUxaQqUR4TfYC28VJFDwjFQNp9QFyKJw7FSMxMh83va0ag&s=aU5Dw6A8r_QtPT6BL2wX6gIwg8ubXnQFBxPxuCaTDpA&e=>. You are receiving this because you authored the thread.Message ID: ***@***.***>

[marcmengel]

Can it check that the provided file has proper permissions and complain and stop if not? Let the user fix their problems instead of trying to do this for them, as it penalizes other users. Andrei

We tried that, but users complained they were

• making tarfiles of areas whose permissions they did not have permission to modify (with tardir:) , or
• using tarfiles provided by others,
  and they found that behavior unacceptable.

Also, decompressing and reading the whole tarfile to check the permissions on everything is not significantly faster than copying it and modifying it.

Just how big is this tarfile you're sending?

[marcmengel]

Also, why are you asking that a file already in /pnfs/mu2e/resilient be re-copied to a dropbox: location, when it is already in resilient? Just leave the dropbox: off of the front and use it where it is...

[gaponenko]

I had to run a test job and knew of a code file I could use, it happened to be in resilient. Let's not get sidetracked by this specific example, the issue is that jobsub_submt unconditionally performs moves that are not needed for well-formed inputs, and those moves inconvenience some users. Code tarballs can be large, in the Mu2e case many hundred MB, so rewriting them is slow. And this re-writing was probably the root reason for the TMPDIR issue (#545). Without re-writing code tarballs the rest of the temp files would fit into /tmp without problems. I think it will be best to streamline job submission by removing the tarball rewrite from jobsub_submit. It can have an option like --redo-my-tarball-just-in-case, but this "extra help" functionality may be better placed into a separate script to prepare a job submission tarball. Andrei

…

________________________________________ From: Marc Mengel ***@***.***> Sent: Thursday, February 29, 2024 9:25 AM To: fermitools/jobsub_lite Cc: Andrei Gaponenko; Author Subject: Re: [fermitools/jobsub_lite] wasteful re-compression (Issue #543) [EXTERNAL] – This message is from an external sender Also, why are you asking that a file already in /pnfs/mu2e/resilient be re-copied to a dropbox: location, when it is already in resilient? Just leave the dropbox: off of the front and use it where it is... — Reply to this email directly, view it on GitHub<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_fermitools_jobsub-5Flite_issues_543-23issuecomment-2D1971377531&d=DwMFaQ&c=gRgGjJ3BkIsb5y6s49QqsA&r=O47fc5vzDTR2V_gla4Ub0Q&m=xZH8ySU4mBtdnsCO1j-5w-5pPoK8OEvno-GSK-AjefCx54bob0iEfjC_GKAPcHgj&s=E0kIKVEYoV9aIBM43UrpwyPFhGGt5mK7-3no35cLWhA&e=>, or unsubscribe<https://urldefense.proofpoint.com/v2/url?u=https-3A__github.com_notifications_unsubscribe-2Dauth_AAXVCGR5SIREZLETDFSTS7LYV5D5JAVCNFSM6AAAAABD64YGM2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTSNZRGM3TONJTGE&d=DwMFaQ&c=gRgGjJ3BkIsb5y6s49QqsA&r=O47fc5vzDTR2V_gla4Ub0Q&m=xZH8ySU4mBtdnsCO1j-5w-5pPoK8OEvno-GSK-AjefCx54bob0iEfjC_GKAPcHgj&s=ksVuRwZtR7bdHlAu9K7oSirlAzYxmMjaS8M0IcykQ3k&e=>. You are receiving this because you authored the thread.Message ID: ***@***.***>