Hmbown/CodeWhale · error · PersistenceBacklogError

baseline_observation.retained_queued_requests exceeds accept

Error message

baseline_observation.retained_queued_requests exceeds accepted_requests

What it means

Raised by validate_budget() when baseline_observation.retained_queued_requests is greater than baseline_observation.accepted_requests (validated earlier as exactly 128). A queue cannot retain more requests than were accepted, so retained > accepted signals duplicated queue entries or corrupted counters.

Source

Thrown at scripts/check-persistence-backlog-budget.py:291

    if baseline_accepted != FIXTURE["requests_attempted"]:
        raise PersistenceBacklogError(
            "baseline_observation.accepted_requests must equal requests_attempted"
        )
    baseline_applied = non_negative_integer(
        baseline.get("applied_version"), "baseline_observation.applied_version"
    )
    if baseline_applied != FIXTURE["expected_applied_version"]:
        raise PersistenceBacklogError(
            "baseline_observation.applied_version must be the final sent version"
        )
    baseline_retained = baseline["retained_queued_requests"]
    baseline_payload = baseline["estimated_retained_payload_bytes"]
    if baseline_retained == 0 or baseline_payload == 0:
        raise PersistenceBacklogError(
            "baseline_observation must retain the final request and payload"
        )
    if baseline_retained > baseline_accepted:
        raise PersistenceBacklogError(
            "baseline_observation.retained_queued_requests exceeds accepted_requests"
        )
    if baseline_payload < baseline_retained * FIXTURE["content_bytes_per_request"]:
        raise PersistenceBacklogError(
            "baseline_observation payload is smaller than frozen retained content"
        )
    provenance = baseline.get("provenance")
    if not isinstance(provenance, dict):
        raise PersistenceBacklogError("baseline_observation needs provenance")
    if provenance.get("platform") != "macos":
        raise PersistenceBacklogError("baseline provenance platform must be macos")
    if not isinstance(provenance.get("source_sha"), str) or not SOURCE_SHA_PATTERN.fullmatch(
        provenance["source_sha"]
    ):
        raise PersistenceBacklogError("baseline provenance needs an exact source SHA")
    if provenance.get("source_dirty") is not False:
        raise PersistenceBacklogError("baseline provenance must identify a clean source tree")
    for field, prefix in (("rustc_version", "rustc "), ("cargo_version", "cargo ")):

View on GitHub (pinned to 8880682c63)

Solutions

1. If the value was hand-edited, correct retained_queued_requests to the observed value <= accepted_requests
2. If the real measurement shows duplication, fix the retention path so each request is retained once - do not raise accepted_requests instead
3. Add a regression test asserting retained <= accepted on receipts

Example fix

// before (budget.json)
"baseline_observation": {
  "accepted_requests": 128,
  "retained_queued_requests": 256, ... }

// after: deduplicated retention
"baseline_observation": {
  "accepted_requests": 128,
  "retained_queued_requests": 128, ... }

def baseline_retained_bounded(budget: dict) -> bool:
    b = budget.get("baseline_observation", {})
    return b.get("retained_queued_requests", 0) <= b.get("accepted_requests", 0)

When it happens

Trigger: retained_queued_requests of 129+ alongside accepted_requests 128 - a hand-edit typo, or a real duplication bug where the persistence queue stores the same snapshot more than once.

Common situations: Duplication regressions like the v0.9.5 journal repair (legacy messages plus the canonical journal double-counting retained work); hand-edit typos; counting applied plus queued requests as retained.

Related errors

• baseline_observation.{field} exceeds its ceiling
• baseline_observation payload is smaller than frozen retained
• rss_during_delta_bytes is inconsistent
• rss_after_delta_bytes is inconsistent
• budget document_kind must be {BUDGET_KIND}

AI-assisted analysis of Hmbown/CodeWhale@8880682c63 (2026-08-16). Data as JSON: /api/errors/2c630e3723ef5290. Report an issue: GitHub.