Skip to content

Dead-letter diagnostics lose history on re-index (firstSeenAt reset to the rebuild) #46

Description

@norgejbb-byte

Expected

A dead-letter's firstSeenAt reflects when the failure was first observed and
survives an index rebuild, so operators can distinguish long-standing failures
from new ones and see the trend.

Observed (bytes)

Public surface: GET https://community-production-9ab1.up.railway.app/api/dacs/status
(OpenAPI: params deadLetterLimit ≤ 100, locator ^stor-[0-9a-f]{40}$; response
indexer.deadLetterDiagnostics {scope:"storage-read", total, byCode, byKind, query, returned, hasMore, items[]}).

At ?deadLetterLimit=100: total: 56, byCode: {"STORAGE_UNREADABLE": 56}, and
all 56 items[].firstSeenAt fall within a single ~4.7-hour window
(2026-07-26T13:44:32.219Z2026-07-26T18:25:05.033Z) — i.e. clustered at the
index rebuild, not spread across the period the failures actually span. Each item
also shows retryState: "exhausted".

The failure population is not new: Community #20 (2026-07-13 era) reports prior
dead-letter counts of 38 and 41; the count has since climbed to 56.
Those earlier entries predate the rebuild, yet every firstSeenAt now reads as
the rebuild time — the original first-observed history was lost on re-index.

Repro

curl -s 'https://community-production-9ab1.up.railway.app/api/dacs/status?deadLetterLimit=100' \
 | jq '.indexer.deadLetterDiagnostics
        | {total, byCode,
           firstSeenMin: ([.items[].firstSeenAt]|min),
           firstSeenMax: ([.items[].firstSeenAt]|max)}'
# → total 56; byCode {STORAGE_UNREADABLE:56}; min/max firstSeenAt inside one ~4.7h window
# compare against the counts (38, 41) in Community #20 to see the pre-rebuild history

Cross-links

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions