Hi, hossman wrote: > > : We index using 4 processes that read from a queue of documents. Each > process > : send one document at a time to the /update handler. > > Hmmm.. then you should have a message from the LogUpdateProcessorFactory > for every individual "add" command that was recieved ... did you crunch > those to see if anything odd popped up (ie: duplicated IDs) > > what did the "start commit" log messages look like? > > (FWIW: I have no hunches as to what caused that behavior, i'm just > scrounging for more data) >
A quick check did show me a couple of duplicates, but if I understand correctly, even if two different process send the same document, the last one should update the previous. If I send the same documents 10 times, in the end, it should only be in my index once, no? The "start commit" message is always: start commit(optimize=false,waitFlush=true,waitSearcher=true,expungeDeletes=false) hossman wrote: > > : Yes, I double checked that no delete occur. Since that indexation, I > : re-index the same set of documents twice and we always end up with 7725 > : documents, but it did not show that ~10000 documents count that we saw > the > : first time. But the difference between the first indexation and the > others > : was that the first time, the indexation last a couple of hours because > the > : documents were not always accessible in our document queue. The others > > Hmmm... what exactly does yout indexing code do when the documents aren't > available? ... and what happens if you forcibly commit in the middle of > reindexing (to see some of those counts again) > If no document is available, the threads are sleeping. If a commit is send manually during the re-indexation, it just commit what has been sent to the index so far. I will redo the test with the same documents and in the same conditions as in our first indexation to see if the counts will be the same again. Again, thanks a lot for your help. -- View this message in context: http://old.nabble.com/Documents-disappearing-tp27659047p27794641.html Sent from the Solr - User mailing list archive at Nabble.com.