Hi,

I'm observing a behavior in Flink 1.20.0 that contradicts the
documentation. The JavaDoc for OperatorCoordinator#start() says:

This method is called once at the beginning, before any other methods.


However, I see that OperatorCoordinator#resetToCheckpoint() is called even
before start(), as part of the object graph creation.

The start() method and the JavaDoc in question were introduced in commit
41b6bfa (FLINK-15099). The code that calls resetToCheckpoint() before
start() was introduced in commit 52f7b9d (FLINK-16177).

Does this mean that this JavaDoc is obsolete and should be corrected? What
are the actual guarantees between the order of the calls to these two
methods?

Thank you.

-- 
Sergei Morozov

Reply via email to