2019-02-18: Development Planning¶
Kacper, Mike L, Craig, Mike H, Kyle, Adam, Tim
Development planning¶
- Updates
- Discussions
- Discussion of Kubernetes storage drivers
- NFS model seems suitable – but with girderfs
- We are happy to use the existing Pod lifecycle
- We just need storage to move with the Pod
- Need to handle credentials
- MH:
- https://itnext.io/how-to-create-a-custom-persistent-volume-plugin-in-kubernetes-via-flexvolume-part-1-f6d9d966e123
- There are basically two methods: Container Storage Interface - this looks like the GO based solution and documented at https://github.com/container-storage-interface/spec/blob/master/spec.md#getplugininfo
- Then there’s FlexVolume, which has a POSIX command line interface
- Discussion of reproducibility round-tripping
- Ana’s use case: A researcher has published a Tale, the claim is challenged, someone else tries to run it to see it. Share back their Tale, share it for comment.
- For example, some package has changed (apt package, etc) that causes different results.
- Example: Registered replication
- KK: Sharing/ACLs will partially address this scenario, but we lack a mechanism for feedback and this is largely out of scope.
- TM: What happens to my original results if I share a Tale with someone?
- CW: Note discussion of results file as part of provenance discussion.
- Discussion of Home directory as volatile based on the sustainability of the project.
- Discussion of Kubernetes storage drivers
Updates¶
- Kacper:
- A lot fixes for an upcoming v0.6 release
- Preparing for EAB Review on Thursday
- Will release rc2 today/tommorow
- Craig:
- Development tasks
- Deploy staging.wholetale.org on Chameleon
- Test plan issue template
- Smoke test issue template
- Release testing and PR review
- Non-development tasks
- EAB preparation
- RR workshop coordination
- TACC PY4 planning (remote execution at TACC and UChicago)
- Todo
- Screencasts
- Development tasks
- Mike H:
- Experimenting with deploying girder on k8s
- Investigated k8s’ storage drivers
- Mike L:
- Testing for v0.6, following the test plan
- Wired up DELETE /dataset endpoint (#386)
- Many, many tiny PRs in prep for v0.6
- Chrome thoroughly tested
- More Firefox testing still needed
- Safari / Edge testing would be nice too
- Adam:
- no updates
- Tim:
- What are the prospects for reproducibility round-tripping? I.e., support for researcher B who has trouble reproducing the work of researcher A (that was published as a Tale) to share his or her attempt to reproduce A’s results (as another Tale derived from the first one) back with researcher A, so that A can help spot the problem and possibly fix or update the original Tale so that it really is reproducible (or reproducible again).