Hi Amrit, thanks for the reply. I shut down all of the nodes on the source cluster after the buffer was disabled, and there was no change to the tlogs.
On Tue, Apr 17, 2018 at 12:20 PM, Amrit Sarkar <sarkaramr...@gmail.com> wrote: > Chris, > > After disabling the buffer on source, kind shut down all the nodes of > source cluster first and then start them again. The tlogs will be removed > accordingly. BTW CDCR doesn't abide by 100 numRecordsToKeep or 10 numTlogs. > > Amrit Sarkar > Search Engineer > Lucidworks, Inc. > 415-589-9269 > www.lucidworks.com > Twitter http://twitter.com/lucidworks > LinkedIn: https://www.linkedin.com/in/sarkaramrit2 > Medium: https://medium.com/@sarkaramrit2 > > On Tue, Apr 17, 2018 at 8:58 PM, Susheel Kumar <susheel2...@gmail.com> > wrote: > > > DISABLEBUFFER on source cluster would solve this problem. > > > > On Tue, Apr 17, 2018 at 9:29 AM, Chris Troullis <cptroul...@gmail.com> > > wrote: > > > > > Hi, > > > > > > We are attempting to use CDCR with solr 7.2.1 and are experiencing odd > > > behavior with transaction logs. My understanding is that by default, > solr > > > will keep a maximum of 10 tlog files or 100 records in the tlogs. I > > assume > > > that with CDCR, the records will not be removed from the tlogs until it > > has > > > been confirmed that they have been replicated to the other cluster. > > > However, even when replication has finished and the CDCR queue sizes > are > > 0, > > > we are still seeing large numbers (50+) and large sizes (over a GB) of > > > tlogs sitting on the nodes. > > > > > > We are hard committing once per minute. > > > > > > Doing a lot of reading on the mailing list, I see that a lot of people > > were > > > pointing to buffering being enabled as the cause for some of these > > > transaction log issues. However, we have disabled buffering on both the > > > source and target clusters, and are still seeing the issues. > > > > > > Also, while some of our indexes replicate very rapidly (millions of > > > documents in minutes), other smaller indexes are crawling. If we > restart > > > CDCR on the nodes then it finishes almost instantly. > > > > > > Any thoughts on these behaviors? > > > > > > Thanks, > > > > > > Chris > > > > > >