I believe this is to be expected. The documents in each shard replica are
kept in sync, however, the underlying index files will get out of sync.
When a node is restarted, it starts syncing the any index files that are
out of sync. See here for more details:
It would be nice to be able to lock the cluster during upgrades, which
would still allow searches, but any index operations would be rejected
which would avoid needing to sync over all the files.
On Thursday, October 4, 2012 12:22:28 PM UTC-6, Jae wrote:
I am updating elasticsearch cluster from 0.19.9 to 0.19.10. Minor version
update is not the main reason of cluster update but I installed
elasticsearch as a daemon tool process.
I am updating one by one. After restarting elasitcsearch, I am waiting
until the cluster health status is turning to green from yellow but it's
taking too long. I've waited more than 20 minutes, but it's still yellow.
Is there any guideline about cluster update?