perf(pxe): derive sender tagging finalization from log blocks

[nchamo]

Motivation

We are trying to reduce the RPC calls PXE makes to nodes. We are starting with one of the most common workflows, which runs every time a user sends a message.

How it works now

Before sending a message for a secret, PXE has to pick the next tagging index to use. It queries the node for the logs of a window of tags, starting just above the highest finalized index. It then calls the node again once per pending tx in that window to read its receipt, which includes txs that no log surfaced, such as the ones this PXE sent earlier. A receipt reporting the tx as finalized is what lets us advance the highest finalized index.

The change

A log already carries the block it was mined in, so we no longer need to ask the node about its tx. We compare that block against our locally synced finalized tip instead, and a window whose logs settle every pending tx now costs a single query.

We still fetch receipts for the txs the logs cannot settle. A tx absent from the window needs one to say whether it is still in the mempool or was dropped. A tx whose highest tracked index is missing onchain partially reverted, and only its tx effect tells us which of its tags survived.

One behavior changes. Logs are queried per secret, so finalizing from them advances only the secret being synced, while a receipt covers the whole tx. A tx that also carried logs for another secret keeps its entry there until that secret syncs on its own.

Metrics

Ran the client flow benchmarks on the base commit and on this branch. Every call we save is a receipt fetch. getTxReceipt drops from 35 to 23, which takes the total from 226 to 214, and no other RPC method moves.

Per flow, the savings land on the transfers, which are the flows that emit private logs. The private_fpc variants gain the most because paying through a private FPC means a second private transfer.

Flow	RPC calls	Round trips	getTxReceipt
transfer_1_recursions+private_fpc	38 → 32	24 → 22	11 → 5
transfer_0_recursions+private_fpc	39 → 35	27 → 25	9 → 5
transfer_1_recursions+sponsored_fpc	23 → 21	17 → 15	5 → 3
amm_add_liquidity_1_recursions	39 → 39	27 → 27	4 → 4
token_bridge_claim_private	26 → 26	14 → 14	1 → 1
storage_proof_7_layers	11 → 11	7 → 7	1 → 1

The unchanged flows are the control: they show the change does nothing to workflows that do not send private logs.

[Thunkar]

Awesome!

[AztecBot]

Flakey Tests

🤖 says: This CI run detected 1 tests that failed, but were tolerated due to a .test_patterns.yml entry.

\033FLAKED\033 (8;;http://ci.aztec-labs.com/b40bc059fbaa00f4�b40bc059fbaa00f48;;�): yarn-project/kv-store/scripts/run_test.sh src/sqlite-opfs/internal/ordered-binary-browser.test.ts (1s) (code: 0)