Extended stall during checkpoint with large number of dirty tables under WiredTiger
WiredTiger changes for MongoDB 3.1.9
WT_SESSION.create does a full-scan of the main table
This would be a highly desirable feature to handle smooth upgrades of production systems for us.
Message: Merge pull request #2189 from wiredtiger/index-create-lsm3
WT-147: Dynamic Index creation. Use bulk=unordered
Message: SERVER-19954 Don't scan tracked handles during checkpoints.
The change in WT-147 to allow a session to lock the same handle exclusive multiple times makes this unnecessary.
I can see that now every I time I call WT_SESSION::create("index:my_table:my_index") for full-scan of the main table is performed to fill index table.
We used to call WT_SESSION::create() for all our tables and indices on each start and now it's not free anymore.
Dmitri Shubin, I'm going to leave WT-147 resolved, and open a new ticket, WT-2295 to handle this question. (In general, new tickets that link to older tickets are easier in our process, sorry!)
Well, actually I didn't want to create a case because probably it's desired and expected behaviour.
Just was surprised a bit...