...

• Correctness tests with induced failures
• Compare checkpoints times under backpressure with current state
• Compare throughput under backpressure with current state
• Compare progress under backpressure with frequently induced failures
• Compare throughput with no checkpointing

Rejected Alternatives

There were several alternatives for specific design decisions:

Support of rescaling in non-keyed operators:

...

For the first implementation of the persistence layer of in-flight data:

• Writing onto DFS/S3 would introduce a new system dependency.
• Implementing DFS on TaskManagers would require too much effort for now.
• Using book keeper seems promising but would also add a new system dependency.