Closed aksestok closed 1 month ago
@IlyaFaer take a look please
@aksestok, @rudolfix, after all, you may be right. For all cases, Kafka returns offset -1001
. That's a nasty constant, which is supposed to be an invalid offset.
Though it's marked invalid, Kafka actually accepts our offsets. But! In case subscribe()
called, Kafka reverts offsets we just assigned. And even if I explicitly commit offsets after assign()
, subscribe()
still reverts to the previous offset (though -1001
is supposed to use the last committed offset, and we just committed it).
Seems to me the problem is in these -1001
(which comes from backend - in code we have normal digits). All the tests are passing without subscribe()
(but they do not pass anymore with subscribe()
🫤, I have two failures), so I think we can merge it.
The craziest is that before and after subscribe()
, Kafka returns the same assignments and positions. Though under the hood they are not.
All the tests are passing without subscribe() (but they do not pass anymore with subscribe() 🫤, I have two failures)
@IlyaFaer do we miss an essential test? extract the messages, drop
the pipeline (to simulate state rollback) and run it again making sure the same messages are back?
@rudolfix, done:
Tell us what you do here
Short description
See #450.
Don't subscribe after assigning partitions and offsets.
Related Issues
Additional Context