Understanding a triggered assertion

[etiennemlb]

What could possibly lead to errno being set to 7 at the following line:

mpifileutils/src/common/mfu_io.c

Line 1051 in 0a4c530

MFU_ABORT(-1, "Failed to write file %s errno=%d (%s)",

This is triggered when making a Tio sized copy of many small files on a recent Luster FS.

[ofaaland]

@etiennemlb we have observed a similar issue in production but failed to reproduce it in test so far. When we saw it:

• lustre 2.12.9_3llnl clients
• lustre-2.14.0_17.llnl servers, backed by zfs-2.1
• on file creation (not write)
• using a PFL default layout, with a DoM component
• while copying from an older Lustre FS to a newer one via dsync

On those same, the sysadmins removed the DoM component from the default PFL layout, making no other changes to either clients or servers, and file creates started working again.

We observed the same failure (before removing DoM) and success (after removing DoM) with the "touch" utility so it was not an mpifileutils issue.

We haven't created an upstream Whamcloud issue against Lustre because we can't reproduce ot understand what's going on, but believe it was a Lustre issue in the end.

I don't know if that's related to your issue or not, but thought it was worth mentioning.