Different author values in blocks table when running Indexer on two identical servers

[TranHuyHiep]

Hi team,

I found an unexpected behavior when running the Midnight Indexer on two different servers.

Both servers:

• run the same version of the Indexer
• use identical configs (same environment, same chain, same docker compose profile, same settings)
• connect to the same network
• are syncing without errors

However, when I compare the Postgres databases, I see that the blocks table has different author values for the same block height between the two servers.

Specifically:

• hash,protocol_version, parent_hash and timestamp fields same data
• but the author field is different between the two machines

❓ My question:

• Why can this field differ?
• Where exactly does the author value come from (which part of the node / protocol / metadata)?
• Is there documentation explaining this field and how it is derived?

Thanks for your guidance!

[hseeberger]

Thanks for reporting, @TranHuyHiep.

The author is determined from information present in the block header and from the set of authorities which is periodically fetched from the Node. You can find details here (https://github.com/midnightntwrk/midnight-indexer/blob/main/chain-indexer/src/infra/subxt_node.rs#L228) and here (https://github.com/midnightntwrk/midnight-indexer/blob/main/chain-indexer/src/infra/subxt_node.rs#L513).

As the information present in the block header (the slot index) is deterministic, the only explanation for the behavior you observed is different sets of authorities being used. Authorities are fetched "initially" or on receiving a "new session" event; the latter again is deterministic.

The "initial" fetching of authorities happens for the first block that is handled after the highest stored block. If you start with a fresh Indexer, that is the genesis block and if you start at some non-empty state it is a different one. This could explain what you have observed. I will try to reproduce.

[hseeberger]

@TranHuyHiep, why did you close this issue? So far I was not able to reproduce.

[TranHuyHiep]

@hseeberger Oh sorry, I closed the issue because the explanation made sense to me, and I assumed the behavior was expected.

If you need the issue to remain open for reproduction or further investigation, feel free to reopen it, I’m happy to provide any additional details if needed.

[hseeberger]

@TranHuyHiep, are you sure (sorry for asking) you ended up with different authors? If so, for which/how many blocks and really for the same Node/chain? Is this reproducible?

[TranHuyHiep]

@hseeberger Yes, I’m sure the author values were different.
I compared both databases at the same block height, and only the author field differed while all other fields (hash, timestamp, parent_hash, protocol version…) were identical.

I suspect it happened because the two Indexers did not start from the same initial block, meaning each one fetched a different initial authority set.

I can reproduce it by running one Indexer from an empty database (starting at genesis) and another Indexer starting from a later height.

I can provide specific block heights and database snapshots if needed.

[hseeberger]

That would be great, because I tried the same and was not successful.

[rongurlavi]

Is it related to #317 ?
@hseeberger

[hseeberger]

@rongurlavi, that should have been fixed.