com.atproto.sync.getLatestCommit returns 501, blocking relay crawl

[simnaut]

Update (re-scoped): This is a spec-conformance gap, not the cause of repos failing to index. The actual indexing/freezing problem is missing prevData in firehose #commit events — see #169. Details below.

Summary

com.atproto.sync.getLatestCommit is not implemented. Requests fall through the registered routes to the catch-all XRPC proxy (app.all("/xrpc/*", …) in packages/pds/src/index.ts) and return 501 MethodNotImplemented. It's a standard sync endpoint and should return 200.

Reproduction

curl 'https://<pds-host>/xrpc/com.atproto.sync.getLatestCommit?did=<did>'
# → HTTP 501 {"error":"MethodNotImplemented","message":"Method Not Implemented"}

Reproduces on any cirrus deployment (verified on pds.simnaut.io and mk.pds.mk.gg).

Not the indexing blocker

This issue originally suspected getLatestCommit of blocking relay indexing. Further investigation disproved that:

• The reference relay (indigo) and rsky-relay build repo state from the firehose and do not call getLatestCommit during crawl (indigo only serves it; getRepo 3xx-redirects to the source PDS).
• relay.feeds.blue — which surfaced this — runs standard indigo (not a custom implementation) and never calls getLatestCommit during crawl. Its repos freeze because of a different bug: cirrus omits the prevData field from #commit events, so strict indigo verification rejects them ("missing prevData field"). That's Firehose #commit omits prevData, breaking inductive verification on strict relays #169, the real federation fix.

So no relay's indexing depends on this endpoint; it's purely conformance.

Notes

All the data the endpoint needs is already available via AccountDurableObject.rpcGetRepoStatus() (returns { did, head, rev }), which getRepoStatus/listRepos already use. The fix is small — see #168.

[simnaut]

Related observation: this also leaves already-indexed repos stuck. feeds.blue reports @ascorbic's repo as active: true, but its view is stale by ~3.5 months:

• relay.feeds.blue → rev: 3me7toj67qs2c (TID timestamp: 2026-02-06 21:24 UTC)
• mk.pds.mk.gg → rev: 3mmk5tdw2mc27 (TID timestamp: 2026-05-23 19:14 UTC, current)

The relay is active but pinned to a commit from early February; the PDS has advanced past it since.

com.atproto.sync.getLatestCommit on mk.pds.mk.gg returns 501 — the same endpoint gap described above, which blocks the relay's crawl from re-fetching the repo.

[simnaut]

Update / re-scoping. Further investigation shows getLatestCommit is not what blocks relay indexing. The actual cause of cirrus repos going stale on relays is a separate firehose bug: #commit events omit the prevData field, so relays running strict commit validation (indigo with LenientSyncValidation off) reject every commit after the first. Filed as #169.

Notably, the relay in this issue (relay.feeds.blue) runs standard indigo, not a custom implementation, and never calls getLatestCommit during crawl — it consumes the firehose and fails at commit verification ("missing prevData field"), which is #169, not this. The operator confirmed they run with LenientSyncValidation off.

This issue still stands as a minor spec-conformance fix — com.atproto.sync.getLatestCommit is a standard sync endpoint and should return 200, not 501 — and PR #168 implements it. But it is conformance/cleanup, not the indexing fix. #169 is the one that restores federation to strict relays.