How does Flink decide when to take a checkpoint?

Question

I'd like to understand what determines when checkpoints are taken. How does this relate to the checkpointing interval?

Answer 1

To a first approximation, the Checkpoint Coordinator (part of the Job Manager) uses the checkpoint interval to determine when to start a new checkpoint. This interval is passed when you enable checkpointing, e.g., here is it set wait for 10 seconds between checkpoints:

env.enableCheckpointing(10000L);

or it can also be set via execution.checkpointing.interval.

However, the Checkpoint Coordinator will wait however long is necessary to avoid violating either the setting for execution.checkpointing.max-concurrent-checkpoints or for execution.checkpointing.min-pause (which defines how much time must elapse between the completion of one checkpoint and the start of the next one).