Add new Healer operation #40
Conversation
pkalever
force-pushed
the
healer
branch
3 times, most recently
from
39765f5
to
e70074f
Compare
Yes. The |
Thanks @nixpanic, a few minutes later I had realized that I need to add the proto file too, so pushed the proto file too, PTAL |
There was a problem hiding this comment.
You will need to add a new capability in https://github.com/csi-addons/spec/tree/main/identity#getcapabilities too.
| // active (staged/published) volume. | ||
| service HealerNode { | ||
| // NodeHealer is a procedure that gets called on the CSI NodePlugin. | ||
| rpc NodeHealer (NodeHealerRequest) |
There was a problem hiding this comment.
Maybe call this HealVolume?
I would expect a way to check if a volume needs healing. The current NodeHealer call seems to do that, but also is expected to do the healing?
It would probably be useful to have a check, and only perform healing if the check returns that it is needed. This then can provide better feedback and tracking of actions that were performed. It might even be useful to fallback to some other recovery/healing mechanism in case volume-healing failed (extension for HealNode which might do a reboot or inform medik8s or similar?)
There was a problem hiding this comment.
@nixpanic
NodeHealer() is called on CSI NodePlugin, so it's up to SP to implement/perform healing or not, right? if so why do we need a flag/check?
I'm probably missing something? Is this flag set as part of the request or response?
- If it is part of request then who will be setting it because we don't have any CRDs here.
- If it is part of the response, may be based on the response CSI addon might want to do a reboot or similar?
Could you please make it more clear?
Also cannot we detect healing failed from the message string which is part of the response?
There was a problem hiding this comment.
I would consider two procedures:
GetHealth()returning the state of the volume, possibly with a selection of recovery options (VolumeHealingorNodeHealingrequired)HealVolume()doing the actual healing, if it can heal the volume at all- the CSI-Addons implementation for the CO can then implement a
HealNodefunction (with Kubernetes medik8s might be an option)
This makes it easier to follow the actions, and improve the reporting to users (Kubernetes events maybe?).
CRDs are Kubernetes specific, so such definitions should go in the kubernetes-csi-addons repository instead. There probably should be a way for the CO to configure the checking interval, maybe even per volume. When using Kubernetes, an annotation on the PVC might be suitable for that (depending on the number of options).
Not only the |
Healer Operation is used both for healing the volume and checking for health of the volume, it is upto the CSI driver whether to implement the healer mechanism or just live with health checks. Signed-off-by: Prasanna Kumar Kalever <[email protected]>
Change to the build scripts Signed-off-by: Prasanna Kumar Kalever <[email protected]>
Fixes: csi-addons#20 Signed-off-by: Prasanna Kumar Kalever <[email protected]>
Healer Operation is used both for healing the volume and checking for
health of the volume, it is up to the CSI driver whether to implement the
healer mechanism or just live with health checks.
Fixes: #20
Signed-off-by: Prasanna Kumar Kalever <[email protected]>
NOTE: I will add the proto auto gen files once the review is finalized