Monorepo for Tangled tangled.org

bobbin/hydrant: account moved to a new PDS stays deactivated #4771

Open chadtmiller.com
Open opened by chadtmiller.com

After I moved my account (did:plc:bcgltzqazw5tb6k2g3ttenbj) to a new PDS, Tangled showed it as deactivated for a day: my profile was a 404 and sign-in failed with AccountDeactivated, while my new PDS and Bluesky's relay both said it was active. Hydrant seems to have dropped the new PDS's #account active=true because its PLC lookup still named the old PDS, and nothing it does afterwards brings an account back. Calling activateAccount again on the new PDS fixed it.

Claude Opus 5.5

Timeline (from Bluesky's relay, via Jetstream) #

UTC Event
2026-11-14 23:54:30 old PDS (pds.chad-53c.workers.dev): #account active=false status=deactivated
2026-11-14 23:57:14 PLC op sets #atproto_pds to https://monolith.chadtmiller.com
2026-11-14 23:58:17 new PDS: #account active=true, #identity
2026-11-16 00:00:48 new PDS: #identity (handle re-asserted). Profile still 404
2026-11-16 00:06:37 new PDS: #account active=true (activateAccount again). Profile 200

What I think happens #

From reading core@cff614de and hydrant@ee54e51b:

  1. Bobbin asks Hydrant for com.atproto.sync.getRepoStatus and treats active:false as gone (bobbin/crates/resolver/src/identity.rs, fetch_upstream / inactive_on_hydrant).
  2. Hydrant connects to PDSes directly and enforces host authority (src/ingest/relay/worker.rs, check_host_authority). With the default PLC_URL it resolves through plc.wtf first. If that mirror hadn't seen the PLC op by 23:58:17, 63 seconds after it, the new PDS's #account active=true is dropped as message rejected: wrong host.
  3. After that, nothing puts the account back. handle_identity (src/ingest/relay/handlers.rs) refreshes the doc and can set authority_identity_moved, but never touches repo_state.active. Only handle_account does. So once the doc resolves to the new PDS, Hydrant still keeps the deactivation it took from the old one until the new PDS happens to send another #account.

Step 2 is my inference: I can't see your Hydrant's logs. Steps 1 and 3 come straight from the code.

Suggested fixes (any one would have saved this) #

  • When Hydrant learns a DID's PDS moved (WasStale { identity_moved: true }, or an authorized #identity from a new host) and the repo is held inactive, check the new PDS's getRepoStatus and take its answer.
  • Don't drop a WrongHost #account / #identity straight away. Park it and retry after re-resolving from plc.directory itself, since a mirror can lag behind a migration's PLC op.
  • In Bobbin, when Hydrant says inactive, check the PDS's own getRepoStatus before treating the account as gone.

Workaround for anyone else hitting this #

Have the new PDS call com.atproto.server.activateAccount again (for example, goat account activate, which needs a full session, not an app password). This sends a fresh #account active=true, which Hydrant accepts now that the doc names the new host. updateHandle alone isn't enough.

Sign up or Sign in to add to the discussion.