We are running Solr 5.5.4 and Sitecore 8.2.3. Our collection for the web index occasionally has multiple versions of an item in it. This is causing problems with some of our older search queries that only expect 1 or no documents returned for an item, which is expected in the web database, since we are not doing any A/B content testing.
I found a nice Solr query to run, to see if there are any duplicated items: https://[solr]:8983/solr/sitecore_web_index/select?facet.field=_group&facet=on&indent=on&q=_language:en&wt=json&facet.mincount=2&facet.limit=2147483647&rows=0
When we rebuild the index, it rebuilds where this query returns no facet_fields items. But after a few days of publishes, a few items creep into this list.
We haven't found another event that seems to be correlated to when these occur. The one thing we have noted is that the older version of the item gets an _indextimestamp value of 10-90 seconds after the new one's _indextimestamp value. So it almost seems like some event is occurring to update/recreate an old item after the new one gets created.
We do use workflow for publishing content. What process is it that removes the old version of an item from the web database?
I've scoured the logs and don't see any logged errors around the times where the new or old items are being created.
I'm not sure where to look next.