Comment # 12 on bug 1063638 from Wenruo Qu

(In reply to Oliver Kurz from comment #0)
> ## Observation
> 
> +++ This bug was initially created as a clone of Bug #1017461 +++
> 
> The first part of fixes have been done in bug 1017461 but it seems the btrfs
> maintenance tasks can still have a significant impact on system
> responsiveness. Running the btrfs maintenance jobs, e.g.

IMHO the problem is quota with balance. Unfortunately it's a known bug and at
least I don't have a clear plan to fix.

Would you please try do a balance with quota enabled, nothing else and check if
the responsiveness get any improvement?

If quota + balance has acceptable responsiveness, then at least it's not ab
urgent problem for qgourp.

(In reply to Andre Guenther from comment #9)
> After hard reset i could not mount at all.

This is the real problem, and in fact much more serious than the performance
problem IMHO.
This, and some recent reports in mail list, suggest btrfs is not as safe as we
though for power loss.

The whole concept of btrfs metadata CoW is, as long as your superblock is
updated correctly or not updated at all, whatever happened shouldn't damage
your fs. (metadata should always be fine, CoW data is also fine while nocowed
data is damaged).

All problems caused by hard reset imply a serious problem we should dig
further.
I would start investigate this by introducing new runtime selftest first.
But the problem seems not easy to fix any time soon.