qsfs issues from trying on the grid

[OmarElawady]

• ygg is slow, and the hook zstor --store calls are blocking, so writing 10GB gets divided into 16MB blocks and a huge number of processes is created causing zstor daemon to be OOMed. Using the public ipv6 didn't cause this problem for 10GB, but since writing to the fs is faster than the network (needs verification), there must be a number for which this issue happens.
• There's no retrying or queuing of zstor done in the hooks, which means failed zstor stores are lost forever.
• zstor daemon when SIGKILLed fails to start because the socket file exists and it doesn't unlink it (easy to fix but should be considered with the issue above).
• when zstor died, zdbfs was blocked and any operation on the filesystem would block (even df -h on the host). Storaged apparently does something that causes it to block in this case causing EVERYTHING to block (provisiond, ...).
• When the volume becomes full, creating the delete mark file fails with Disk quota exceeded (working on a fix)

[muhamadazmy]

Why this issue is open on zos. It's all qsfs/zstor issues as far as i can tell

[maxux]

From what I know, zstor act like a daemon and have internal queue of process like that (in a single process), you should not have tons of running process running.

@LeeSmet ?