2 Node cluster questions


#1

Hello

Appreciate this topic has lots of topics already, but I have a few specific (but general) questions;

Use case:
Web logs, cron logs, other logs into an elastic stack so devs can look at logs, spot problems etc.
100% availability not required, but would be annoying to reindex all the logs again from scratch.
Sort of unrelated - also want somewhere to run jenkins and tests

Current situation
1 Machine running full elastic stack (elastic/kibana/logstash) & Jenkins, selenium and some other tools.
Jenkins is slow when elastic is doing stuff, and visa versa, so everything is just generally slow :frowning:
I have 4 indexes, all now setup as 1 shard (and one replia), for a daily period. (might change this to weekly for some indexes as each shard doesn't hold much data (50mb).

Proposal
As this is just for internal dev, don't have a super huge budget :money_with_wings:
I am thinking about setting up 2 machines, like the pretty picture on the add failover guide

Machine 1

  • elastic master & data
  • kibana
  • logstash
  • jenkins slave

Machine 2

  • elastic data (replica)
  • jenkins master
  • other testing tools

Fully appreciate it is not recommended to run a 2 node elastic cluster, but can't really justify another machine just to make elastic happy, so;

1 master to stop split brain situation. We're not too fussed if master dies for a little bit, so long as we can get it back manually in a 'reasonable' amount of time

Question 1:
Say machine 1 blows up. Can I do the following, and will it work as I think it will?
I promote machine 2 to a master temporarily (replica shard happy to turn into primaries?), then setup a new machine 1 as a data only node. Data is then replicated across. I then stop machine 2, make machine 1 a master, reset machine 2 back to data only. Everyone is happy, machine 1 is master, machine 2 is data replica and don't have to reindex all logs?

Question 2:
Kibana - I set this to talk to machine 1. Will elastic on machine 1 work out that there is a data node it can talk to as so share out some of the work?

Question 3:
Rolling upgrades - I just upgrade master as per docs, then data only. Everything happy?
Full cluster restart - I stop master and data, upgrade master, then data. Everything happy?

Many thanks!


(Luiz Santos) #2

Hi @cosmo,

I will answer your question based on your statement that this is not a production bulletproof cluster.

Question 1:
Instead of doing all these changes if a node fails I would let both nodes be master/data nodes and set minimum_master_nodes: 2. If one node fails elasticsearch cluster will reject indexing documents. Then you can either take some action to have 2 nodes again or set minimum_master_nodes: 1 temporarily until the second node is up again. This can be done through REST API.

If you do the way you are proposing I think it will work too, but it's more error-prone.

Question 2:
In this case, all request will arrive in machine 1 but elasticsearch will distribute processing across both nodes. A better solution would be load balancing but as you said you don't want to add more nodes....

Question 3:
Both rolling upgrade and full restart will work :slight_smile:

Hope it helps.

Cheers,
Luiz Santos


#3

Thanks Luiz!

I'll look at having them both Master/data nodes then.


(system) #4

This topic was automatically closed 28 days after the last reply. New replies are no longer allowed.