[draft/WIP] Install as Gardener extension

[ppalucki]

Install CRI-RM as Gardener extension using ControllerDeployment/ControllerRegistation as generic Extension of type "cri-resource-manager"

TODO:

I. PoC/research phase

• [x] 1. P ~No operator, just using ManagedResource with shoot namespace is hardcoded so it won't install automatically on every new shoot~ (replaced by proper operator)
• [x] 2. M Replace example configmaps with "installation daemonset" similar to how gvisor is installed here
• [x] 3. P Fully fledged operator based on gardener/extension pkg framework (actuator with migrate/hibernate + operation annotation + status management + finalizers),
• [x] 4. P Prepare "extension" image + example controller registration/deployment + automatic generation of those
  • [ ] 4a. may require to add support for GardenLinux (to have a dockerfile to build cri-rm for gardenlinux)
• [x] 5. M Replace downloading Debian package with "installation" image
• [x] 6. M Make installation more reliable
  • [x] distro agnostic - installation should use static binary + own systemd unit files (so far, exiting debian package and binaries assume e.g. GCC_2_23 version available)
  • [x] check services statuses before restarting kubelet
• [ ] 7. M Uninstallation flow
  • [ ] 7a. When shoot is deleted
  • [ ] 7b. (low priority) when extension is disabled/removed from podspec

Bugs:

• [ ] Race condition when restarting kubelet can causes to cluster become irresponsible
• [ ] If resync is set to low value - there is race between reconciliation and delete handlers - my object is both deleted/created and the same time - and deleting of shoot is stuck

II. Quality

• [x] 7. M Health check for operator
• [ ] 8. M Advanced Self-healing installation: health check e.g. use static pod as health check and undo installation if something fails
• [x] 9. P/M Proper project structure: Makefile, cmd/pkg, hacks/examples, licenses (adhere to convention of other gardener extensions)
• [ ] 10. P/M Unit tests
• [ ] 11. M Integration tests - with envtest NewShootFramwork cluster
• [ ] 12. P E2E tests + workload e.g. cloud setup on AWS

III. Advanced/Extra features

• [ ] 13. P Image vector support (overriding extension and installation image)
• [ ] 14. M Support for cri-resource-manager advanced features:
  • [ ] Annotation webhook - to pass pod spec resource information to CRI-RM
  • [ ] Node agent - to support dynamic policies

Check the README.md for further details (how to actually run this within kind local cluster.

[codecov-commenter]

Codecov Report

Merging #760 (8606110) into master (0cb0ad5) will increase coverage by 0.00%. The diff coverage is n/a.

:exclamation: Current head 8606110 differs from pull request most recent head 237bb70. Consider uploading reports for the commit 237bb70 to get more accurate results

@@           Coverage Diff           @@
##           master     #760   +/-   ##
=======================================
  Coverage   37.51%   37.52%           
=======================================
  Files          55       55           
  Lines        8108     8105    -3     
=======================================
- Hits         3042     3041    -1     
+ Misses       4771     4770    -1     
+ Partials      295      294    -1     

Impacted Files	Coverage Δ
...rce-manager/policy/builtin/topology-aware/pools.go	62.05% <0.00%> (+0.15%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 0cb0ad5...237bb70. Read the comment docs.

[ppalucki]

Gardener extension was moved to external repo (with proper project structure/testing and so on) here: https://github.com/intel/gardener-extension-cri-resmgr

[ppalucki]

Gardener extension was moved to external repo (with proper project structure/testing and so on) here: https://github.com/intel/gardener-extension-cri-resmgr