what todo with broken cachedev?

Hi list,

assuming a cached device mylv from /dev/sda1 with cachepool from /dev/sdb1 (ssd i.e.):

If now the ssd crashes and the cachemode is "writethrough" i think, there is no data loss for my lv's, right?

But then mylv runs without a cache device although the device is set in metadata.

The command lvreduce cant remove the cache-attribut. vgreduce *could* remove it, right? But only if there is no usage bei the cachepool-device anymore.

Is there a solution for that?

Dne 9.8.2014 19:25, Oliver Rath napsal(a):

Quoted 8 lines

Hi list,

assuming a cached device mylv from /dev/sda1 with cachepool from /dev/sdb1 (ssd i.e.):

If now the ssd crashes and the cachemode is "writethrough" i think, there is no data loss for my lv's, right?

Quoted 8 lines

But then mylv runs without a cache device although the device is set in metadata.

The command lvreduce cant remove the cache-attribut. vgreduce *could* remove it, right? But only if there is no usage bei the cachepool-device anymore.

Is there a solution for that?

Cache support is still experimental - so please open bugzillas for your issues found on bugzilla.redhat.com
(if you are not using Fedora/RHEL - use Fedora Rawhide for your reports)

I assume your problem is likely related to still missing repair support for cached volumes.

You would need to use cache utils (cache_check, cache_repair, cache_dump...) to fix broken metadata.

Also, it would be nice when using "writethrough" to have the option of abandoning the cache - making it a normal LV again. If there is no way to get the SSD back, there is probably no point in fixing. Operation might have to be forced if "writeback".