Hi everyone,
I'd like to start a discussion on FLIP-572 [1].
When a Flink job using exactly-once KafkaSink fails and does not recover,
Kafka transactions can remain in the ONGOING state, blocking all downstream
read_committed consumers at the Last Stable Offset until the broker timeout
expires.
There is currently no built-in tooling to resolve this, Kafka's own
kafka-transactions.sh cannot commit Flink transactions since that requires
Flink-specific internals.
This FLIP proposes a standalone CLI tool, that allows operators to abort or
commit lingering transactions without a running Flink cluster.
Looking forward to your feedback.

Kind regards,
Aleksandr Savonin

[1]
https://cwiki.apache.org/confluence/display/FLINK/FLIP-572%3A+Introduce+Flink-Kafka+Transactions+Management+Tool

Reply via email to