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
{{ message }}
Repository navigation
Make resource pipeline dispatch recoverable after redelivery and session-lock loss #5236
The API saves operation state and sends deployment messages in separate actions. If processing stops after a message is accepted but before that fact is saved, recovery can send the same step again. Overlapping workers can also replace newer status with stale data.
Related changes are combined into fewer PRs for review. Earlier PRs and test results remain as historical evidence.
Keep open until message dispatch and recovery have a complete implementation and integration evidence.
Position
Deferred investigation. No complete recovery design or implementation is claimed.
Source finding
This is a code-derived risk, not the confirmed cause of the certificate failure in #5233.
Partial protection
PR #5253 adds uninstall callback checkpoints and stale-callback guards. It does not guarantee one message delivery or exclude concurrent stale writers.
Scope
Define recovery before sending, after message acceptance and before acknowledgement.
Prevent repeated destructive actions, lost deployment requests and stale status writes.
Cover install, upgrade, uninstall, custom actions and cascades.
Remaining work
Design durable message dispatch and repeat-safe processing across Cosmos DB and Service Bus.
Add integration tests for process interruption, redelivery and overlapping workers after session-lock loss.
Document queue compatibility and upgrade requirements for the chosen design.
Activity
Run each resource processor step once after a session lock loss (micr…
Send the first step message again after an API stop (microsoft#5236)
Refuse an upgrade after an uninstall and fail a step without storage …
Let the first run replace a failure that the API wrote (microsoft#5236)
Check the session after the lease wait and let an upgrade repair a lo…
Fail an unsent first step without a change to other resources (micros…
Count a later request on the first child of a pending cascade (micros…
Add tests that fail if the earlier fixes go back (microsoft#5236, mic…
Fail an upgrade when the Porter installation lookup fails (microsoft#…
Keep a newer execution record after the lease expired (microsoft#5236)
Keep the results of earlier steps in the execution record (microsoft#…
Describe when a copy of a step runs Porter again (microsoft#5236)
Write a result over a running record of the same step (microsoft#5236)
Clear the execution ID of a step for a status without one (microsoft#…
Fail an unsent step only if the operation did not change after the cl…
Describe the known limits of the execution lease and the resend (micr…
Describe two runs after a worker stops during a Porter action (micros…
Name the status that a workspace keeps after a failed resend (microso…
Add an appended array item only once if the API prepares a step again (…