: Well, in the same processes I am using a jdbc connection to get all the : relative paths to the documents I want to index, then I parse the documents : to plain text using tones of open source libraries like POI, PFDBox : etc.(which might account for java2d) then I add them to the index and commit : every 2000 documents.
Since nothing in your threaddumps refers to solrj or solr (either as the current method, or in the call stack suggesting that it's code called by Solr(J)), there's really no indication that the problem is even remotely solr related. i suspect that if you commented out all of hte code where you use SolrJ so that you still did all of the parsing and then just wrote the resulting data to /dev/null you would probably still see this behavior -- perhaps one of the other libraries you are using has some semantics you aren't obeying (ie: a parser that must be used single threaded, or an object that must be closed so that it can reset some static state) that is acusing this problem only after some time has elapsed (or on particular permutations of data) -Hoss
