Details
Improvement
Description
we've observed when indexing a large repository, that over time a large number of index requests fill up the queue, consuming both memory and disk processing. If there is not enough memory allocated, the server can eventually restart, and the indexing will not complete, meaning the queue will fill up again on restart.
This is not a memory leak - if the queue completes all operates as normal. The problem is the queue not being processed quickly enough.
Issue Links
| This issue supercedes: | ||||
| MRM-1058 | Error 500 Unable to load class for JSP |
|
|
|
The problem was that each request re-scanned the index to see if the document was an addition or an update. As the size of the index grew, each indexing task became larger, causing the queue to grow exponentially.
With the fix in place, the queue should now not grow above 5-10 items when operating normally, and should complete a full scan in about 1/3rd of the time.