You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When the operator starts handling a CR (i.e. AppDefinition, Session, Workspace), it marks the resource as HANDLING. If the operator is killed (e.g. by Kubernetes due to resources or an update), the resource is never set to HANDLED.
Currently, resources in HANDLING are never processed again to prevent handling resources that caused unexpected crashes,
However, unexpected crashes can be caught (and should already after #200).
Describe the solution you'd like
When a resource with status HANDLING is processed again, the operator needs to consider that encountering an already completed step does not mean that handling was finished. Instead, the operator should only execute the missing steps.
Describe alternatives you've considered
None, eventually we need to be safe against operators being stopped while handling a resource.
Cluster provider
No response
Additional information
No response
The text was updated successfully, but these errors were encountered:
lucas-koehler
changed the title
Continue resource handling when it was cancelled unexpectedly before
Continue resource handling after being cancelled before
Aug 2, 2023
Adds support to the operator to handle AppDefinitions that are in
HANDLING state. This happens when the operator was unexpectedly shut
down while handling the AppDefinition before.
No logic changes in creating Kubernetes resources are necessary because
the handling was already idempotent.
Contributed on behalf of STMicroelectronics
The Workspace handling already makes sure steps are not executed twice.
Thus, we can simply try handling again
Contributed on behalf of STMicroelectronics
Is your feature request related to a problem? Please describe.
Follow up of #200.
When the operator starts handling a CR (i.e. AppDefinition, Session, Workspace), it marks the resource as
HANDLING
. If the operator is killed (e.g. by Kubernetes due to resources or an update), the resource is never set toHANDLED
.Currently, resources in
HANDLING
are never processed again to prevent handling resources that caused unexpected crashes,However, unexpected crashes can be caught (and should already after #200).
Describe the solution you'd like
When a resource with status
HANDLING
is processed again, the operator needs to consider that encountering an already completed step does not mean that handling was finished. Instead, the operator should only execute the missing steps.Describe alternatives you've considered
None, eventually we need to be safe against operators being stopped while handling a resource.
Cluster provider
No response
Additional information
No response
The text was updated successfully, but these errors were encountered: