If a replica shard falls for a long time and then is up again, does this
replica overwritten by the replica created in another node or does being
updated only with the missing parts.
If the new replica overrides the old, does it create a traffic issue, if
the replica is very big?
If a replica shard falls for a long time and then is up again, does
this replica overwritten by the replica created in another node or
does being updated only with the missing parts.
Depends on what data is lost. ES can avoid recopying data if there
are Lucene segments intact on the replica.
If the new replica overrides the old, does it create a traffic
issue, if the replica is very big?
ES will throttle the number of replications per node to attempt to
not saturate the network. If the shard is large, though, it can
indeed take a while to copy. This is almost entirely dependent on
the network. ES adds negligible overhead here.
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.