Comment 14 for bug 1349711

Hm, I'm not sure I can give a thorough description since I don't understand enough about the exact workload myself. It is a fairly arbitrary workload generated by our users.

In the end, it boils down to creating, reading and writing many (~20,000) sqlite files of size 16kb - 12GB across many folders and doing random read/write IO to them. The directory structure is that all 20,000 files live inside a directory of one root directory, like so:

/path/00001/file.sqlite
/path/00002/file.sqlite

etc.

The 500GB volume has approximately 275GB of such files. When copying /path/00001/file.sqlite to /path/00001/file.sqlite.new (and so on) with `cat` twice in parallel (via xargs -P2 as in #11), the volume eventually (after multiple hours) hangs. If the copying is resumed from the last file successfully copied before the hang, the hang onset is very rapid.