Skip to content

Test/chain busy e31 reserve release - #19

Merged
yilmaztayfun merged 2 commits into
masterfrom
test/chain-busy-e31-reserve-release
Sep 16, 2026
Merged

yilmaztayfun merged 2 commits into
masterfrom
test/chain-busy-e31-reserve-release

Conversation

@yilmaztayfun

Copy link
Copy Markdown
Contributor

No description provided.

yilmaztayfun and others added 2 commits September 13, 2026 00:19
…chain reserve

Extends ChainBusyAcceptTests with the other half of the accept-time reserve:
what happens when the forward it was taken for never succeeds.

'auto-leaf-to-waiting' exists in the leaf's definition but is not available in
'leaf-waiting'. The root has no such transition, so it accepts the request,
marks the chain Busy down to the leaf and forwards — and the leaf rejects it
with Transition:100020 (Validation). The post-commit failure policy treats that
as the client's error and declines to fault, so neither settlement nor fault
runs. Before the runtime fix nothing undid the reservation and every level
stayed Busy permanently, with no incident and no retry path.

The test pins three things: the observed status returns to Active so the client
can act again; the ancestors stay Busy because each legitimately holds an open
SubFlow correlation and must not be settled mid-subflow; and the chain still
works afterwards, reaching root-done.

Differential run (2026-09-12, local stack): with the runtime fix stashed and the
hosts rebuilt the test fails at 31s with the chain still Busy; restored, the
whole ChainBusy suite passes 15/15.

Council 2026-09-08-parent-notification-mechanism, P0 (E31).

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
The POST .../instances/events endpoint had no integration test of any kind —
recorded as a named gap by agent council 2026-09-11-route-trace-coverage.

A test that posted to the endpoint directly would prove the mapping script and
nothing about the wiring that actually delivers production traffic: the topic,
the subscription route, the CloudEvent unwrap and the triggerType 3 gate. So both
messages go through the broker, and the test never calls the runtime's API to
drive the flow. The instance is created by an event and advanced by an event,
correlated by business key alone — no instance id is ever published.

The flow follows the documented configuration rather than an invented one: a
workflow-level event mapping for action=start, a triggerType 3 transition with
its own mapping for action=transition, and two Dapr Subscriptions whose routes
carry action and transitionKey. The approve subscription uses sync=true so the
delivery blocks until the pipeline reaches a rest point.

Three things are pinned, including the one a state-only assertion would miss: the
transition must run with the MAPPED body, not merely move the state. And an
approval naming no active instance must be acked rather than retried — the
endpoint answers the broker's protocol, so "nothing matched" must not look like a
failure or the broker redelivers a message this runtime can never satisfy.

The README records the trap that cost time here: Dapr reads declarative
subscriptions only at startup, so a subscription added while the sidecar is
running sends messages nowhere, with no error and no log.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
@yilmaztayfun yilmaztayfun self-assigned this Sep 16, 2026
@yilmaztayfun
yilmaztayfun requested review from a team September 16, 2026 07:45
@coderabbitai

coderabbitai Bot commented Sep 16, 2026

Copy link
Copy Markdown

Important

  • 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: 1a942bcd-4a7b-4fae-990d-93571181325f


Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

Comment @coderabbitai help to get the list of available commands.

@yilmaztayfun
yilmaztayfun merged commit a8ae081 into master Sep 16, 2026
2 of 3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant