getLatestCommit on Relay returns 500 for (some?) deleted accounts

[mackuba]

https://bsky.network/xrpc/com.atproto.sync.getLatestCommit?did=did:plc:pkmfz5soq2swsvbhvjekb36g
https://bsky.network/xrpc/com.atproto.sync.getLatestCommit?did=did:plc:ravrgr24ltd5oxft2mha6chc
https://bsky.network/xrpc/com.atproto.sync.getLatestCommit?did=did:plc:cw2oqmrjzpfaezhs6pp3bksy

HTTP/2 500 
vary: Origin
date: Fri, 15 Mar 2024 22:17:37 GMT
content-length: 0
strict-transport-security: max-age=63072000

The first one is a DID from an independent PDS where the owner (IIRC) has manually added a plc_tombstone operation to PLC because they thought it was needed. The second one is also a repo on another independent PDS, the repo doesn't seem to exist on the PDS anymore.

The third was a DID on a Bluesky PDS, so I'm assuming this might actually be happening for all deleted repos?

[mackuba]

More examples:

did:plc:icjkfdhae2trwtx3ikmv62ii (maitake.us-west.host.bsky.network)
did:plc:5voyzxlltvyvfzw5umluvfu5 (bsky.gdgd.tokyo)
did:plc:g6klig5oawtvtqb7qnuyhhwb (bsky.gdgd.tokyo)

[bnewbold]

Thanks for reporting these! Sorry to leave these lingering.

We are planning to iterate on how identity and account lifecycle events are transmitted over the repo event stream ("firehose"), most likely deprecating the #handle and #tombstone events to use just #identity and #account. When we do that, the relay will be updated to have slightly different internal account state tracking. I think that we'll probably wait to debug these accounts until we do that update, and then ensure/test that tombstoned accounts are handled consistently.