Wrong write hit/miss statistics

[katlapinka]

Description

When running the same write workload twice, we expect that after the first run there will be 100% misses and after the second there should be 50% misses and 50% hits - in reality we have a really small amount of hits (~5%) and other writes are counted as misses.

[mmichal10]

@katlapinka does the problem occur on v22.6.3? I'm curious if the problem has been introduced by the io forward mechanism in v24.9. Could you also share your configuration and the script you used to reproduce the problem?

[katlapinka]

The problem does not occur on v22.6.3

FIO config:

echo "[global]
size=0.5*cache_size
rw=randwrite
ioengine=libaio
blocksize=65536
iodepth=1
group_reporting
[core]
filename=/dev/cas1-1" | fio --eta=always --output-format=json --output=fio_run_debug -

Doesn't matter if I/O is sequential or random write.

Cache config:

Cache Id                  1
Cache Size                3424 [4KiB Blocks] / 0.01 [GiB]
Cache Device              /dev/vdb1
Exported Object           -
Core Devices              1
Inactive Core Devices     0
Write Policy              wb
Cleaning Policy           nop
Promotion Policy          always
Cache line size           64 [KiB]
Metadata Memory Footprint 18.0 [MiB]
Dirty for                 64 [s] / 1 [m] 4 [s]
Status                    Running

Sequential cutoff is disabled.