On Wed, Jan 23, 2013 at 08:37:10PM +0800, Nitin Kshirsagar wrote:
Hello Kent,
I have setup bcache and run fio data verification test on write back and write through caches.
The fio tests are passed, however I have found following issues while using bcache.
Issue1: Cache is not created as per user specified options
--------------------------------------------------------------------------------------------------------
Steps:
1.Create cache by specifying mode writeback and cache replacement policy as fifo
[root@annu bcache]# make-bcache --cache /dev/sdc --bdev /dev/sdd --writeback --cache_replacement_policy=fifo
UUID: e25f2840-f02b-46af-81e7-28948b2737cc
Set UUID: 68da5b89-1e87-457a-80c7-2c822737f969
nbuckets: 2048
block_size: 1
bucket_size: 1024
nr_in_set: 1
nr_this_dev: 0
first_bucket: 1
UUID: a3ce52e6-631b-4c74-afa2-9f8b0088c7f4
Set UUID: 68da5b89-1e87-457a-80c7-2c822737f969
nbuckets: 20480
block_size: 1
bucket_size: 1024
nr_in_set: 1
nr_this_dev: 0
first_bucket: 1
[root@annu bcache]# echo /dev/sdc > /sys/fs/bcache/register
[root@annu bcache]# echo /dev/sdd > /sys/fs/bcache/register
2. Cache mode should be "writeback" instead of "writethrough"
[root@annu bcache]# cat /sys/block/bcache2/bcache/cache_mode
[writethrough] writeback writearound none
[root@annu bcache]# cat /sys/block/bcache2/bcache/writeback_running
1
3. Cache policy should be "fifo" instead of "lru"
[root@annu ~]# cat /sys/block/bcache2/bcache/cache/cache0/cache_replacement_policy
[lru] fifo random
These two are... oddities/documentation bugs with make-bcache. Options
apply to devices you specify _after_ the option, not before - it was
implemented this way so you could format cache devices and backing
devices at the same time - and really, the other options should be
parsed differently but I haven't spent much time on make-bcache.
Patches welcome :)
Issue2: Cache dirty data value should not be negative.
-------------------------------------------------------------------------------------------------------
Steps:
1.Create cache by specifying mode writeback and cache replacement policy as fifo
2.To make bcache devices known to the kernel
[root@annu bcache]# echo /dev/sdc > /sys/fs/bcache/register
[root@annu bcache]# echo /dev/sdd > /sys/fs/bcache/register
3.Create FS on cache /dev/bcacheN and mount in directory
4.Create Data set by using fio or dd on mount point.
5. Change cache node from "writethrough" to "writeback"
[root@annu ~]# echo writeback > /sys/block/bcache2/bcache/cache_mode
[root@annu ~]# cat /sys/block/bcache2/bcache/cache_mode
writethrough [writeback] writearound none
6.Check cache dirty data should not be negative value
[root@annu ~]# cat /sys/block/bcache2/bcache/dirty_data
-9.4M
Known bug, but fairly harmless one. That value is only used for the pd
controller that adjusts the background writeback rate - and in practice
when there's dirty data in the cache it shouldn't be _too_ far off.
It's on the todo list, but it'll be a bit before I get to it.