My PDS Stopped Syncing on January 23rd.

[hackdefendr]

All,

I am developing a web frontend to Bluesky called SkyDeck Live. I'm only stating this, because at some point on January 23rd while testing my code, the Bluesky Network lost my PDS subscription. I am not finding anything in my PDS logs to tell me why this happened.

• My PDS (hackdefendr.com) stopped syncing to the relay around January 23rd
• com.atproto.sync.requestCrawl returns success but relay still shows "account not found"
• All PDS sync endpoints are accessible and working

I badly need help debugging this issues and hopefully re-establishing the relay subscription.

Jeff

[bnewbold]

It looks like the Bluesky relay knows of this PDS instance, and has a recorded cursor value from it of 11,416. That means at some point this PDS hostname emitted an event with that sequence number.

However, when I connect to the PDS just now, the most recent firehose event I can see has a sequence of 3,073.

The relay will ignore old events (to prevent re-processing, loops, and other problems), and the PDS is probably returning a "future cursor" error if the relay tries to connect with a high sequence.

Do you know what might have caused this situation?

In terms of un-breaking things, ideally there would be a way to reset the PDS to a higher cursor if there was an operational problem (eg, corrupt disk). Alternatively, you could emit a bunch of dummy events to move the cursor forward (eg, create a dummy account and do about 8000 dummy repo commit operations). Or, in the most disruptive case, create a new PDS with a new hostname and migrate all the accounts there.

[hackdefendr]

Hi ... thanks for replying.

I have zero clue as to what happened. I am just as baffled as you seem to be. I was hoping it was part of my development testing, but no amount of rolling back to any January 23rd commit worked to solve it. All my testing shows "account not found" from the relay.

Luckily, there is only one account. I tried changing my account name from jeff.hackdefendr.com to just hackdefendr.com and that worked, but it did not re-initiate the connection with the relay. All my posts are stuck on my server, I confirmed this with a wsdump command that should normally never show any data unless there's a problem.

I don't suppose there is a goat cli command that will move the cursor or fake 8000 posts?

[hackdefendr]

Here is a single log from my backend showing the "account not found" error:

Delete post error: XRPCError: Account not found
    at BskyAgent.call (/var/lib/skydeck/live/backend/node_modules/@atproto/xrpc/dist/xrpc-client.js:76:23)
    at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
    at async AppBskyFeedPostRecord.delete (/var/lib/skydeck/live/backend/node_modules/@atproto/api/dist/client/index.js:1185:9)
    at async RateLimiter.execute (file:///var/lib/skydeck/live/backend/src/utils/rateLimiter.js:176:24)
    at async BlueskyService.deletePost (file:///var/lib/skydeck/live/backend/src/services/bluesky.js:377:7)
    at async AuthService.withAutoRefresh (file:///var/lib/skydeck/live/backend/src/services/auth.js:407:14)
    at async file:///var/lib/skydeck/live/backend/src/routes/posts.js:77:18 {
  error: 'AccountNotFound',
  headers: {
    'access-control-allow-origin': '*',
    'cache-control': 'private',
    connection: 'keep-alive',
    'content-length': '57',
    'content-type': 'application/json; charset=utf-8',
    date: 'Sat, 31 Jan 2026 02:08:17 GMT',
    etag: 'W/"39-Aaseh9F4IjdFkVEKnOscIcyUNlk"',
    server: 'nginx/1.22.1',
    vary: 'Authorization, Accept-Encoding',
    'x-powered-by': 'Express'
  },
  success: false,
  status: 403
}

[hackdefendr]

Is it possible to have the cursor reset for my PDS domain without having to create 8000 dummy posts? I'm fine with losing posts, but I just don't want to start over again.