May 19, 2020. 1PM PST
Zoom information: Apache Cassandra Contributor Meeting
Video Link:
Agenda: Post agenda items below
- Update on Kubernetes Operator CEP-2
- Cassandra CI Run-through, next steps, help needed, and Q&A (Mick)
Meeting Minutes
Update on Kubernetes Operator CEP-2
- Cassandra k8s operator SIG meeting Thur
- Some cleanup remaining on doc. This Thursday talking about moving the initiative forward.
- Many operators out there already. Need to sort out among them and get alignment.
- Q: have we talked to people who have been successful running C* on k8s?
- That has been the experience in SIG meetings to date
- Bloomberg and New Relic newly interested
- Q: CEP-2 goals have some overlap with CEP-1. Should we contribute to CEP-1 so non-k8s deployments can take advantage as well?
- There have been some discussions
- Example: running a nodetool command across the cluster, declarative nodetool command issuance
- in the DS sidecar concept, one issue is that they didn’t want to get into dealing with authentication off-node
- Agreed: CEP-2 should use the CEP-1 solution. No duplicated effort.
- Add an item to Thur k8s meeting (?)
- Sidecars: Let a thousand flowers bloom?
- Nobody will stop it, but folks see value in having a reference implementation.
- Sometimes it’s a struggle to find compromise on that sort of thing
- Notion of sidecar is different depending on application (e.g. k8s operator vs. TLP Reaper)
CI/CD (Mick)
- Few messages to dev list outlining the work done on Cassandra Jenkins
- Cassandra CI was meant to establish a permanent record of test results
- Makes it possible to do the scale testing without access to an expensive CircleCI plan.
- Q: should we be using Jenkins CI as well to validate?
- they do different things
- CircleCI does not do cqlshlib tests. Hesitation about adding because some want to remove the tests (effort to run in CircleCI > porting to Python dtest)
- No upgrade dtests
- No jdk11 build/test (somebody working on this)
- No jdk11 test of jdk8 builds
- Do we have to be committers to use Jenkins CI?
- yes, unless you are part of the companies donating infrastructure
- Hadoop apparently does pre-commit builds (maybe more projects)
- Jira annotations?
- Jeff: Hadoop uses https://yetus.apache.org/
- Adding Harry and Fallout to CI process?
- Requires too much resources at this point
- Also not in Jenkins: drivers, upgrade tests.
- Current capacity: can be saturated on busy days
- Is it possible we don’t want pre-commit tests? Should Jenkins be reserved for the longer-running post-commit stuff?
- A few rough edges
- cqlshlib tests are not containerized, and will bomb if running concurrently
- Fielded some ideas about k8s based testing. Also Stupp is working on testing with Gradle Enterprise.
- There is a confluence wiki page with pros/cons of Jenkins and CircleCI
- <a bunch of discussion about scale testing>
- Given that we’re not able to run all on existing infrastructure, it was agreed that the company stakeholders will describe their workload tests and findings ahead of key release milestones.