Our cluster got into really bad shape in production, probably because of too many shards. 45k shards in 5 data nodes.
I have tried a few things and seems to help that includes restarting the node (make sure there is no data loss). Right now, I see about 15k unassigned shards, but these unassigned hsards are recovering so slow that it will probably take days to recover.
I know there is a delay set for allocation of unassigned shards when a node leaves as mentioned here, but that's not helping here.
Is there a way to speed up allocation of unassigned shards with all nodes up? We have 2 replicas for all indices.