On the surface, this doesn't make sense, I'd expect that the tlogs would be roughly the same size on leaders and replicas. Or at least show the same variance.
If you were to guess how much volume in terms of files being fired at the index, how much would you expect in 30 seconds? And does it approximate the size you're seeing in your tlogs (acutally 2x your data transmission rate over 30 seconds). Hard commits with openSearcher=false are actually pretty cheap operations. About all they do is close the currently open segments and truncate the tlog. What happens if you drop it to 10 seconds? Best, Erick On Wed, Apr 2, 2014 at 4:04 AM, Gurfan <htc.ja...@gmail.com> wrote: > Thanks Shawn for the quick reply. > > We are using Solr Cloud version 4.6.1 > > Usually we see higher transaction log on replica. Leader`s tlog size is in > KB`s. We also tried keeping the hard commit(autoCommit) as 20 Sec and > autoSoftCommit as 30 Sec. > > We written a script to monitor the disk usage of tlog directory in 1 Min > interval, also noticed that the logs are purge at a particular time. For > instance: Tlog starts with ~4MB and it increases at some point i.e. > 20MB,50MB,220MB,600MB again it reduces with ~10MB. > > http://searchhub.org/2013/08/23/understanding-transaction-logs-softcommit-and-commit-in-sorlcloud/ > > How can we reduce Tlog size at its lowest, so that our system restart up > time will less. > > Thanks, > --Gurfan > > > > > > > -- > View this message in context: > http://lucene.472066.n3.nabble.com/transaction-log-size-tp4128354p4128547.html > Sent from the Solr - User mailing list archive at Nabble.com.
