State | Parameter | Pre-FLIP | Post-FLIP |
WaitingForResources | resource-stabilization-timeout | - default: 10s
- Trigger: Change event
| - Renamed to submission.resource-stabilization-timeout
|
resource-wait-timeout | - default: 5min
- Trigger: State initialization
- Can be disabled
| - Renamed to submission.resource-wait-timeout
|
Executing | scaling-interval.min | - default: 30s
- “Cooldown”: No change events in that timeframe trigger any rescale
- Trigger: Change event
| - Renamed to executing.cooldown-after-rescaling
- The cooldown period will be disabled (i.e., 0ms) in the WaitingForResources state.
- A slight change in the Executing state: Timeout starts during state initialization.
|
scaling-interval.max | - default: n/a
- Timeout, after which even sufficient resources are enough to trigger rescaling
| - Replaced with executing.resource-stabilization-timeout with default value 60s.
|
min-parallelism-increase | - default: 1
- Minimum increase in parallelism for a job to scale up.
| - The parameter will be deprecated and not used anymore.
- This parameter becomes unnecessary as the number of scales will be reduced by the cooldown period and stabilization timeout.
|
max-delay-for-scale-trigger | - default: n/a
- The maximum time the JM will wait to evaluate previously observed events for rescaling.
- Since Flink 2.0 (not yet released; FLIP-461)
| - Replaced with rescale-trigger.max-delay
- Simple renaming is enough since this parameter will be just introduced with Flink 2.0
- The trigger will be set to
Duration.ZERO for WaitingForResources state because there are no checkpoints used
|
scale-on-failed-checkpoints-count | - default: 2
- The number of failed checkpoints will trigger rescaling even without a completed checkpoint.
- Since Flink 2.0 (not yet released; FLIP-461)
| - Replaced with rescale-trigger.max-checkpoint-failures
- Simple renaming is enough since this parameter will be just introduced with Flink 2.0
|