States condition to be allowed to write into index

I want to bulk-insert into an index.
This index is made of 3 shards (0, 1, 2) and 1 replica distributed over 2 nodes as follow :

shard 0 p
shard 1 r
shard 2 p

N2 :
shard 0 r
shard 1 p
shard 2 r

I loose N2 (because of network issue, or something else), the delay is passed and so shard 1 r on N1 become shard 1 p on N1.

When N2 come back, shard 1 r is reindexed.

This situation seems to lead to a freeze of my bulk-insert (it waits a long time to finally falls with this kind a message "all nodes failed").

So, I would like to know what is the normal behaviour behind this kind of situation ? When a replica's missing, what's going on ?


In order to have a highly available cluster that is able to continue accepting writes if one node is lost you need a minimum of three nodes. If you only have 2 master eligible nodes the loss of one node will mean no master can be elected as this requires a strict majority of master eligible nodes to be available and writes will be blocked as a result even if primary shards are available.

Thank you Christian.

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