Stability of latest Elasticsearch?


After seeing the CVE's for older ES, I figure it's time to maybe upgrade from 1.4.5. The last time I tried an upgrade, it was a disaster...ES couldn't handle a couple service restarts without failing completely with errors. Can anyone attest to the stability of the current ES? Thank you.

(Jörg Prante) #2

Can you be more specific? Why did you do a couple of service restarts? What errors do you mean? What is "failing completely"?


This was a while's what I have from my post a few months ago in this list:

marking and sending shard failed due to [failed recovery]

was the error...ES completely stopped collecting data at that point. I removed all the data and rolled back to 1.4.5, which has been working fine since then. As for the service restarts, there are certain times that, after running a fair amount of time, I'll see random CPU spikes. Doing a:

sudo service elasticsearch restart

fixed that. But running the above a few times resulted in that failure above. Hope fully that explains it. Thank you.

(Jörg Prante) #4

Shard failures can have many reasons. The info you gave is quite sparse, exception messages and scenario description would be useful to get an idea what happened - I never had any shard failures. Maybe the index recovery could not get back to work because you have custom analyzers or plugins.

Blind restarts never fix error conditions, they only confuse the cluster.

Be sure to check for breaking changes, stop indexing, flush the indices to empty the translog, and follow the steps in

All Elasticsearch 1.4.5 and up use Lucene 4.10.4, so from data stability perspective, I think there is no reason to be concerned about data getting corrupted. You nevertheless always need to create a backup of your data.


That helps thank you....and yea...sorry about the lack of information. I can tell you I have no additional plugins or custom analyzers installed. This setup is pretty basic at home with not a lot of traffic to it. I'll try my upgrade again and report my findings...thanks again.

(Nik Everett) #6

Full cluster restarts have really helped me on older versions when the master has gone deaf. Its worth keeping in your toolbox, just never using blindly.


Thanks Nik...appreciate it.

(system) #8