kami/correx
1
0
Fork 0
Code Issues Pull Requests 2 Actions Packages Projects Releases Wiki Activity

fix(retry): charge gate budget on any repeated failure fingerprint (#460)

Browse Source 
gateFailureFingerprints held one slot per gate, so the budget only charged
when a failure repeated back to back. Whack-a-mole (fix A breaks B, fix B
breaks A) alternates two fingerprints forever, every round read as progress,
and the per-gate budget never triggered. Track the full set seen per gate and
charge when the current fingerprint is a repeat.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
This commit is contained in:
claude
2026-08-02 14:12:47 +04:00
parent d69cb12ce9
commit 1ef845ed0f
5 changed files with 40 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files
testing/kernel/src/test/kotlin/DefaultRetryCoordinatorTest.kt
+26 -5
View File
@@ -179,7 +179,7 @@ class DefaultRetryCoordinatorTest {
val policy = RetryPolicy(maxAttempts = 3, backoffMs = 0L)
val state = OrchestrationState(
gateRetryBudgets = mapOf("contract" to 1),
gateFailureFingerprints = mapOf("contract" to FailureFingerprint.of("same-reason")),
gateFailureFingerprints = mapOf("contract" to setOf(FailureFingerprint.of("same-reason"))),
)
val decision = retryCoordinator.decide(
@@ -200,7 +200,7 @@ class DefaultRetryCoordinatorTest {
// changed fingerprint (progress) must still be free and retry.
val state = OrchestrationState(
gateRetryBudgets = mapOf("contract" to 1),
gateFailureFingerprints = mapOf("contract" to "old-reason"),
gateFailureFingerprints = mapOf("contract" to setOf("old-reason")),
)
val decision = retryCoordinator.decide(
@@ -218,7 +218,7 @@ class DefaultRetryCoordinatorTest {
val policy = RetryPolicy(maxAttempts = 2, backoffMs = 0L)
val state = OrchestrationState(
gateRetryBudgets = mapOf("contract" to 2),
gateFailureFingerprints = mapOf("contract" to FailureFingerprint.of("same-reason")),
gateFailureFingerprints = mapOf("contract" to setOf(FailureFingerprint.of("same-reason"))),
)
val decision = retryCoordinator.decide(
@@ -234,7 +234,7 @@ class DefaultRetryCoordinatorTest {
val policy = RetryPolicy(maxAttempts = 1, backoffMs = 0L)
val state = OrchestrationState(
gateRetryBudgets = mapOf("contract" to 1),
gateFailureFingerprints = mapOf("contract" to FailureFingerprint.of("same-reason")),
gateFailureFingerprints = mapOf("contract" to setOf(FailureFingerprint.of("same-reason"))),
)
// "contract" is already exhausted at maxAttempts=1, but "review" has never charged.
@@ -245,12 +245,33 @@ class DefaultRetryCoordinatorTest {
assertEquals(RetryDecision.Retry, decision)
}
@Test
fun `decide charges a whack-a-mole repeat — fingerprint seen earlier but not last round`() = runTest {
val policy = RetryPolicy(maxAttempts = 3, backoffMs = 0L)
// A/B/A: round 3 repeats round 1's failure. The previous fingerprint is B, so a single-slot
// comparison would read this as progress and never charge.
val state = OrchestrationState(
gateRetryBudgets = mapOf("contract" to 1),
gateFailureFingerprints = mapOf(
"contract" to setOf(FailureFingerprint.of("A"), FailureFingerprint.of("B")),
),
)
retryCoordinator.decide(
sessionId, stageId, gate = "contract", failureReason = "A", state = state, policy = policy,
)
val event = eventStore.read(sessionId).single().payload as RetryAttemptedEvent
assertEquals(true, event.charged)
assertEquals(2, event.attemptNumber)
}
@Test
fun `decide honours perGateMaxAttempts override`() = runTest {
val policy = RetryPolicy(maxAttempts = 1, backoffMs = 0L, perGateMaxAttempts = mapOf("review" to 5))
val state = OrchestrationState(
gateRetryBudgets = mapOf("review" to 4),
gateFailureFingerprints = mapOf("review" to "same-reason"),
gateFailureFingerprints = mapOf("review" to setOf(FailureFingerprint.of("same-reason"))),
)
val decision = retryCoordinator.decide(