The million collection test was written years ago and is very outdated on how it gets executed. It merges the latest MongoDB with the latest WiredTiger and compiles them into a binary for the test to execute. This might have suited us when we dropped WiredTiger into MongoDB/master once in several months. Now that we do a drop almost every day, we are practically wasting effort merging the two repositories. Most of the runner script can be done away with and the test will need a complete re-write with the new features in the plan. The test runs with much fewer collections than a million. 19701 at the moment, though enough, to begin with, we will like to have the capability to have more as the testing progresses. Ideally, we would have several runs with a varied amount of collections, so we get an idea of how performance (or extent of the stalls) varies with the number of collections. Hence, Have the capability to run the test on a "configured" amount of collections. Have the capability to run multiple tests with a variation on the test parameters. Eventually we will like the test to become a performance and a qualifying test on dhandle growth, so a means to capture performance relative to the test parameters like the number of collections will be useful.
WT-7336, a new many-collection test was written in order to be able to create workload where stalls happen while gathering statistics.
This ticket is dedicated to:
- the creation of a script that triggers many-collection-test.py. It is possible to start from the script that has been posted by Sulabh in
- The creation of an evergreen test that triggers the new script.