I couldn't find a recent answer for this (but my search skills after 16 hours of 5-to-6-upgrade-work are a little... downscaled). Long story short: I have to move 14 shards from a broken-beyond-repair node, which is booted in rescue mode, to a working node.
Can I simply rsync -a --ignore-existing /var/lib/elasticsearch/nodes/0/indices from the broken node to the working node?
Do I need to stop the working node during the sync?
Turns out: yes, you totally can. Well, better to check the index uuid and select just the missing shards, otherwise you wait quite a while to copy 600GB from one shard via a 1GBit connection.
That was exhaustive... but the numbers look good so far, the garbage collection nightmare with 5.6 and extensive-terms-query caching seems to be resolved, hooray.
Apache, Apache Lucene, Apache Hadoop, Hadoop, HDFS and the yellow elephant
logo are trademarks of the
Apache Software Foundation
in the United States and/or other countries.