[dm-devel] dm_cache BUG() after power loss

[Steinar H. Gunderson]

On Fri, Oct 19, 2018 at 09:03:50AM -0400, Mike Snitzer wrote:
> Please provide the "dmsetup table" line for the cache device if you can.

It's torn down by now, unfortunately. I can create a new volume and set it
up, if it helps you?

> Are you using writeback mode?

No, it's writethrough. I'd love to use writeback, but I've lost too much data
to it, so I don't really trust it anymore. :-/

Is writethrough actually pure writethrough? When recovering, I had repeated
rounds of:

  - Get to initramfs, which tried to set up the cache and failed.
  - fsck the real, underlying volume, which had lots of corruption.
  - Boot from it.
  - Reboot.
  - Repeat to step 1.

In particular, the volume seemed to become corrupt over and over again
(indicating that dm-cache was writing something to it before it failed),
until I took out the lines fromcachetab.

> Given the "block manager: array validator check failed for block 2156"
> error it could easily be that you need to run cache_check and
> cache_repair.

I hadn't heard of these tools before, and they're not mentioned in
device-mapper/cache.txt. (They should probably have been mentioned in the
error message, too.) Where can I find them?

> Joe (cc'd) may have more specific repair guidance for
> you (though Joe is going on vacation.. bad timing).

Repair is way too late anyway; like I wrote in the original message, I had to
just clear out the entire cache volume to get anywhere.

/* Steinar */
-- 
Homepage: https://www.sesse.net/