feat(solana-indexer) PR 13: index on-chain-created orders

[squadgazzz]

Description

Orders placed directly on chain (the demo flow has no orderbook) now get their solana.orders row, which is what makes them visible to the autopilot's open-orders cut. The missing piece was the token mints. The intent only names token accounts, so at flush time the decoder resolves the batch's token accounts through getMultipleAccounts and reads each account's mint out of the SPL layout.

The lookup only trusts accounts owned by a token program with the token-account size, so an intent naming a garbage account cannot smuggle a fake mint into the orders table. When a mint cannot be resolved (not a token account, closed), the order keeps its order_pda row, the intent row is skipped, and the transaction is dead-lettered with reason unresolved_mints, so the replay machinery re-delivers it later. An RPC failure is handled differently: the slot's database write never starts and the last indexed slot does not advance, so a restart re-delivers the whole slot and transient outages lose nothing.

Changes

• The decoder resolves each batch's token accounts to mints before persisting, fn token_account_mint reads the SPL account layout and rejects non-token accounts
• CreatedOrder carries its transaction signature, the dead-letter key for unresolved orders
• apply_order_created writes the solana.orders row (zero fee_amount, the on-chain intent has no fee, and creation_timestamp is indexing time, the stream carries no block time)
• Rpc::new_mock for tests over a canned transport

How to test

New and updated DB tests: the pipeline test now runs through a mocked getMultipleAccounts and asserts the orders row, plus an unresolved-mints case. cargo nextest run -p solana-indexer solana_db --run-ignored ignored-only --test-threads 1, and they run in CI.

Related issues

Resolves BE-207.

[linear-code]

BE-207

[claude]

Claude finished @squadgazzz's task in 2m 35s —— View job

---

PR Review: index on-chain-created orders

Reviewed the diff against origin/main. Overall this is clean, well-documented work — the batched mint resolution, the RPC-failure-vs-unresolved-mint distinction, the idempotent ON CONFLICT DO NOTHING inserts, and the mocked-RPC test coverage are all solid. SQL column/value/bind counts line up correctly, and the serde_json dependency is correctly dev-only.

Two points posted as inline comments (both worth a look, neither blocking):

• decoder.rs token_account_mint — reads the first 32 bytes of any account without confirming it's actually an SPL token account (owner + exact 165-byte length). A non-token account could silently write a garbage mint into solana.orders.
• persistence.rs unresolved-mint skip — the "until a replay fills it in" claim: since the slot still advances here, a restart won't re-deliver the event. Worth confirming a replay path exists or holding the slot back for this case, as is done for the RPC-error path.

Nice touches: distinct mints in the tests so a swapped binding fails, and the unresolved-mints test asserting order_pda is kept while orders is skipped.

• Read the diff against origin/main
• Review decoder changes (mint resolution)
• Review persistence changes (apply_order_created)
• Review rpc / error handling
• Review tests
• Post inline comments + summary