Authors: Aihua Li, George Chen, Yu Li
Status
Current state: "Under Discussion"
Discussion thread: https://lists.apache.org/thread.html/5aac294120d93b418bd6900eeb2416f4f49010241a847830e6ea2ff1@%3Cdev.flink.apache.org%3E
JIRA: here (<- link to https://issues.apache.org/jira/browse/FLINK-XXXX)
...
Page properties | ||||||||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
|
Please keep the discussion on the mailing list rather than commenting on the wiki (wiki discussions get unwieldy fast).
...
- Add test suite for basic operations, and a visible WebUI to check the throughput and latency data, pretty much like our existing flink speed center.
- Add more software and hardware metrics for the benchmark.
- Add test suite for state backend.
- Add test suite for shuffle service.
...
Test Scenarios
The following dimensions of Flink job are taken into account when setting the test scenarios:
Topology | Logical Attributes of Edges | Schedule Mode | Checkpoint Mode |
OneInput | Broadcast | Lazy from Source | ExactlyOnce |
TwoInput | Rescale | Eager | AtLeastOnce |
Rebalance | |||
KeyBy |
There're also other dimensions other than Flink characteristics, including:
- Record size: to check both the processing (records/s) and data (bytes/s) throughput, we will test the 10B, 100B and 1KB record size for each test job.
- Resource for each task: we will use the Flink default settings to cover the most used cases.
- Job Parallelism: we will increase the parallelism to saturate the system until back-pressure.
- Source and Sink: to focus on Flink performance, we generate the source data randomly and use a blackhole consumer as the sink.
Test Job List
The above test scenarios could form 32 test jobs as shown below:
...