Recovery From Missing translog.ckp files with 2.0 Upgrade


(Paul R.) #1

I've run an errant command, and accidentally deleted all */translog/translog.ckp files for my single-node cluster. This occurred during the upgrade process from ES 1.7.3 to ES 2.0.0.

I managed to keep a backup of the actual *.tlog files for all indices, but I don't have a copy of the checkpoint files.

I'm OK with a small amount of dataloss in each index; the alternative seems to be full loss of all of my indices. Is there a way to disable checking the translogs entirely, in order to trigger a recovery attempt? This isn't production, but I'd like to make sure if something similar goes wrong on our production cluster there would be a way to recover.

Any help appreciated.

Cheers,
Paul


(Erik Stephens) #2

A hard crash caused a couple of my indices' translogs to be corrupted. For one index, I was able to remove the translog dir on one node and re-open the index to recover. On the other index, no combination of removing the translog dir on the data nodes helped me recover the index. Sometimes would get a NoSuchFileException, others a FileAlreadyExistsException.

That doesn't help you, Paul, but I'm hoping someone can chime in and confirm what can or can't be done when elasticsearch is unhappy with the translogs.


(system) #3