bq:why does it replicate all the index instead of copying just the newer formed segments
because there's no guarantee that the segments are identical on the nodes that make up a shard. The simplest way to conceptualize this is to consider the autocommit settings on the servers Let's say the hard commits (which close the current segment and open a new one) are all set to 1 minute. The fact that the servers are starting at different times means that the segments on one node will close at different times than another node. And that doesn't even consider the complicated cases of possibly having different segments merged depending on the start/stop pattern on one of the nodes.... Best, Erick On Fri, Aug 16, 2013 at 5:25 AM, Ido Kissos <idokis...@gmail.com> wrote: > Yes, I have erased the tlog in replica 2 and it appears that the the first > replica's tlog was corrupted because of an ungracefull servlet shutdown. > There was no log for it unfortunately, neither the zookeeper log logged > anything about this. Is there a a place I could check in the zookeeper what > exactly happened during this election? > > Partly connected - about transient disk that needs to be free for the > replication after sync failure - why does it replicate all the index > instead of copying just the newer formed segments? That would require much > less space than a full copy, wouldn't it? > Why not making 100 docs for tlog sync configurable? >
