Info
Background:
A primary node writes multiple same-node shard key updates.
A second attempts to replicate, and is behind enough that it decides to batch apply instead of single apply.
Issue:
When the secondary attempts to apply each statement separately, it will see that the document shard key is changing, and will uassert because the opCtx is not in a transaction. It wouldn't be in a transaction, because the batch apply has no transaction associated with itself. This individual uassert leads to an fassert from the replication code.
Fix:
Add a check to !getOpCtx()->writesAreReplicated() in [this uassert here|https://github.com/mongodb/mongo/blob/406e1642a83d8b810bec41ad293c1b1c8a59a6dd/src/mongo/db/exec/update_stage.cpp#L929, so that we don't require the opCtx to have a transaction number if we are in a secondary node.
Top User Comments
xgen-internal-githook commented on Tue, 30 Apr 2019 15:29:08 +0000:
Author:
{'name': 'Blake Oler', 'username': 'BlakeIsBlake', 'email': 'blake.oler@mongodb.com'}
Message: SERVER-40812 Allow replica set secondary node to batch apply same-node document shard key changes
Branch: master
https://github.com/mongodb/mongo/commit/326fd668123f73c5e81ce597d58e40b569374820
