Closed kdave closed 3 years ago
Another data point: right before the zone got reclaimed, 'fi df' showed 'zone_unusable=7.12GiB', so the accounting is correct, but the message in the the log had 0 for unusable.
This one's actually fairly trivial to solve. At the point the message get's printed we have already marked the block group as read only, which migrates all zone_unusable bytes to bytes_readonly and sets it to 0. The solution is to cache block_group->zone_unusable before calling inc_block_group_ro().
I'll send a patch with the message improved and fixed shortly.
Tested, works, thanks.
With the following change
all messages I can find in the syslog contain the value 0%:
The workload is the git checkout that I've been using so far, the nullb0 device is 1G with 128M zone size. The 'df' numbers look normally: