[FLINK-9000] Add automated cluster framework tests

• Type: Improvement
• Status: Closed
• Priority: Critical
• Resolution: Done
• Component/s: Runtime / Coordination, Tests
• Assignee: Unassigned
• Reporter: Till Rohrmann
• Created: 16/Mar/18 12:57
• Updated: 27/Apr/21 18:06
• Resolved: 03/Mar/20 17:38

Description

Similar to the end-to-end tests which are added to the flink-end-to-end-tests module, we should add cluster framework tests which we can run automatically.

One idea how to launch a cluster framework would be to use Terraform to allocate nodes from AWS or GCE. Next we could setup a different cluster framework like Yarn or Mesos on which we run Flink. In order to simulate different failure scenarios a framework like Jepsen could be used.

This issue serves as an umbrella issue to collect all issues which are related to cluster framework tests.

Sub-Tasks

• Cluster test: Run general purpose job with failures with Yarn session: Closed
• Cluster test: Run general purpose job with failures with Mesos: Closed
• Cluster test: Run general purpose job with failures with Yarn and per-job mode: Closed