Actually, I'm surprised that the slave returns the new document and I suspect that there's actually a commit on the master, but no new searcher is being opened.
On replication, the slave copies all _closed_ segments from the master whether or not they have been opened for searching. Hmmm, a little arcane. Here's a long blog on the subject: https://lucidworks.com/2013/08/23/understanding-transaction-logs-softcommit-and-commit-in-sorlcloud/ But... Whenever you hard commit (often configured in solrconfig.xml) you have a choice whether opensearcher=true|false. _IF_ opensearcher=false, the current segment is closed but the docs are not searchable yet. When the slave does a replication, it copies all closed segments and opens a new searcher on them. So here's one possibility: 1> you added some docs on the master but your solrconfig has an autocommit setting that tripped in and has openSearcher=false. This closed all open segments (i.e. the segments with the new docs) 2> the slave replicated the closed segments and opened a new searcher on the index, so it shows the new docs 3> the master still hasn't opened a new searcher so continues to not be able to see the new documents. Is that possible? Erick On Mon, Jul 24, 2017 at 3:04 PM, Stanonik, Ronald <rstano...@ucsd.edu> wrote: > I'm testing replication on solr 5.5.0. > > I set up one master and one slave. > > The index versions match; that is, master(replicable), master(searching), and > slave(searching) are the same. > > I make a change to the index on the master, but do not commit yet. > > As expected, the version master(replicable) changes, but not > master(searching). > > If I "replicate now" on the slave, then slave(searching) matches > master(replicable), which seems wrong because the slave now returns answers > from master(replicable), while the master returns answers from > master(searching). > > Shouldn't the slave continue to return answers from master(searching), so > that master and slave return the same answers? > > What do I not understand? The documentation I found about replication > doesn't seem to explain in depth how the versions are affected by changes and > commit. > > Thanks, > > Ron
