Improve SQL store indices and queries

.github/workflows/test.yml

@@ -203,6 +203,8 @@ jobs:
           save-if: false
       - name: Run miden-bench CLI tests
         run: make test-miden-bench
+      - name: Run the SQL store scaling benchmark
+        run: make bench-store
       - name: Fetch test node binaries
         uses: actions/download-artifact@d3f86a106a0bac45b974a628896c90dbdf5c8093 # v4.3.0
         with:

CHANGELOG.md

@@ -13,6 +13,8 @@
 * [BREAKING][param][rust] `NodeRpcClient` models encrypted submissions: `submit_proven_transaction` now takes `SealedTransactionInputs` instead of `TransactionInputs`, `submit_proven_batch` now takes `Vec<SealedTransactionInputs>` (one per transaction, each sealed against its own transaction ID), and implementations must provide the new `get_transaction_encryption_key` method ([#2341](https://github.com/0xMiden/rust-sdk/pull/2341)).
 * [BREAKING][type][rust] Added the `NoteFilter::ScriptRoots` variant, so exhaustive matches on `NoteFilter` in `Store` implementations must handle it ([#2335](https://github.com/0xMiden/rust-sdk/pull/2335)).
 * [BREAKING][behavior][store] The SQLite base schema now declares an index on `input_notes(script_root)`. This changes the schema fingerprint, so opening a database created before this change fails with `SchemaHashMismatch` and existing stores must be recreated ([#2335](https://github.com/0xMiden/rust-sdk/pull/2335)).
+* [BREAKING][behavior][store] The SQLite base schema now indexes `code_commitment` on `latest_account_headers`, `historical_account_headers` and `foreign_account_code`, the `input_notes` consumption index now leads with `consumer_account_id`, and the `input_notes` state index now carries `nullifier`. This changes the schema fingerprint, so opening a database created before this change fails with `SchemaHashMismatch` and existing stores must be recreated ([#2364](https://github.com/0xMiden/rust-sdk/pull/2364)).
+* [BREAKING][param][rust] `Store::get_input_note_by_offset` is replaced by `Store::get_input_note_after`, which takes an `Option<InputNoteCursor>` identifying the last note read instead of an ordinal offset. Build the cursor for the next call with `InputNoteCursor::from_record`. `Store` implementations must be updated; `InputNoteReader` is unaffected ([#2364](https://github.com/0xMiden/rust-sdk/pull/2364)).
 
 ### Enhancements
 
@@ -21,9 +23,15 @@
 * [FEATURE][rust] `Client::get_consumable_notes(Some(account_id))` now screens only that account instead of screening every tracked account and discarding the rest, so its cost no longer grows with the number of tracked accounts. Added `NoteScreener::get_batch_consumability_for_account` to screen notes against a single account ([#2338](https://github.com/0xMiden/rust-sdk/pull/2338)).
 * [FEATURE][rust] Added the `miden_client::rpc::encryption` module backing encrypted submissions: `TransactionEncryptionKey`, `AttestedTransactionEncryptionKey` (whose `verify` is the only path to a usable key), `ValidatorAttestation`, `NextTransactionEncryptionKey`, `SealedTransactionInputs` and `seal_transaction_inputs`, along with re-exports of the validator DSA key types reachable from this API ([#2341](https://github.com/0xMiden/rust-sdk/pull/2341)).
 * [FEATURE][rust,store] Added `NoteFilter::ScriptRoots` to query input notes by their note script root directly at the store level, without loading and screening unrelated notes. The filter doesn't apply to output notes: querying output notes with it returns an empty list ([#2335](https://github.com/0xMiden/rust-sdk/pull/2335)).
+* [FEATURE][store] Added the `miden-bench store` subcommand, which measures how the `SQLite` store methods scale with the number of notes and accounts and reports the growth between the smallest and the largest size. It seeds its own throwaway databases and needs no node, and `make bench-store` runs the same sweep CI does ([#2364](https://github.com/0xMiden/rust-sdk/pull/2364)).
 * [rust] Added `PartialBlockchainUpdates::block_headers_to_store`, which narrows the staged headers to the ones a sync must persist: those marked as relevant, genesis, and the block at the sync height. `block_headers` still yields all staged headers ([#2297](https://github.com/0xMiden/rust-sdk/pull/2297)).
 * [rust] State sync now authenticates every relevant note block but only persists block headers and MMR authentication nodes for blocks containing notes that remain unspent or that a `NoteObserver` explicitly marks as relevant ([#2297](https://github.com/0xMiden/rust-sdk/pull/2297)).
 
+### Fixes
+
+* [FIX][rust] `InputNoteReader::next` now fails with `ClientError::MissingNoteConsumptionPosition` when the store yields a note that carries no consumption position. The walk cannot advance past such a note, and it previously restarted from the first note on every subsequent call ([#2364](https://github.com/0xMiden/rust-sdk/pull/2364)).
+* [FIX][store] Input notes listed with `NoteFilter::Consumed` now break ties on the details commitment instead of the note ID, so notes consumed by the same transaction come back in a stable order. A note only carries an ID once its metadata is known, which left the previous tie-break undefined for the rest ([#2364](https://github.com/0xMiden/rust-sdk/pull/2364)).
+
 ## 0.16.0-alpha.1 (2026-07-17)
 
 ### Breaking Changes

Makefile

@@ -18,6 +18,10 @@ WARNINGS=RUSTDOCFLAGS="-D warnings"
 
 TEST_MIDEN_NOTE_TRANSPORT_URL?=http://127.0.0.1:57292
 
+# Sizes the SQL store scaling benchmark sweeps over. Kept small enough to run on every PR, and
+# overridable for a deeper local run.
+STORE_BENCH_ARGS?=--notes 1000,10000 --accounts 100,1000 --iterations 5
+
 # --- Linting -------------------------------------------------------------------------------------
 
 .PHONY: clippy
@@ -84,6 +88,12 @@ test-miden-bench: ## Run miden-bench CLI tests
 test-docs: ## Run documentation tests
 	cargo test --doc $(FEATURES_CLIENT)
 
+# --- Benchmarking --------------------------------------------------------------------------------
+
+.PHONY: bench-store
+bench-store: ## Run the SQL store scaling benchmark (no node needed)
+	cargo run --package miden-client-bench --release --locked -- store $(STORE_BENCH_ARGS)
+
 # --- Integration testing -------------------------------------------------------------------------
 
 .PHONY: start-node

bin/miden-bench/README.md

@@ -1,6 +1,6 @@
 # miden-bench
 
-Benchmarking tool for the Miden client library. This binary measures performance of transactions to establish baselines and identify optimization opportunities.
+Benchmarking tool for the Miden client library. This binary measures performance of transactions and of the `SQLite` store to establish baselines and identify optimization opportunities.
 
 ## Installation
 
@@ -66,6 +66,39 @@ The number of storage maps is auto-detected from the account.
 miden-bench --network localhost transaction --account-id 0x...
 ```
 
+### `store`
+
+Benchmarks the `SQLite` store as the database grows.
+
+```bash
+miden-bench store --notes 1000,10000 --accounts 100,1000 --iterations 5
+```
+
+Two tables come out of a run, one per sweep. Each row is a store method, each column a size, and the last column the growth from the first size to the last. That growth is the number to read: a method served by an index stays near `1.00x` while the database grows, and one that falls back to a scan grows with it.
+
+The note sweep seeds three quarters of the notes as consumed by a single account, spread over blocks and transaction orders, and the rest as unspent notes carrying a nullifier. It measures:
+
+- `get_input_notes` for the `Unspent`, `Consumed`, `List`, `Nullifiers` and `ScriptRoots` filters
+- `get_unspent_input_note_nullifiers` and `get_tracked_block_headers`
+- `get_input_note_after` seeded with a cursor near the end of the account's history
+- a full `InputNoteReader` walk, reported as a total and per note
+- `upsert_input_notes` and `apply_state_sync` for a fixed batch of new notes
+
+The account sweep seeds wallets that share one account code, and measures `SqliteStore::new` (which rebuilds the SMT forest on open), `get_account_headers`, `get_account_header` and `prune_account_history`.
+
+Reads run before writes within a size, so every read measures exactly the seeded database.
+
+```bash
+# A deeper sweep. Seeding is linear in the sizes, so the run time is too.
+miden-bench store --notes 10000,100000 --accounts 1000,10000
+```
+
+`make bench-store` runs the same sweep CI does. Override `STORE_BENCH_ARGS` to change the sizes:
+
+```bash
+make bench-store STORE_BENCH_ARGS="--notes 5000,50000 --accounts 500,5000"
+```
+
 ### `import`
 
 Imports an account into the local store. Two mutually exclusive modes:
@@ -196,6 +229,12 @@ miden-bench deploy --maps 3
 - `-r, --reads <N>` - Maximum storage reads per transaction. When total entries exceed this limit, reads are split across multiple transactions per benchmark iteration. Each iteration's time is the sum across all transactions. When omitted, all entries are read in a single transaction.
 - `-i, --iterations <N>` - Number of benchmark iterations (default: 5)
 
+#### Store
+
+- `--notes <N[,N...]>` - Note counts to sweep over (default: `1000,10000`)
+- `--accounts <M[,M...]>` - Account counts to sweep over (default: `100,1000`)
+- `-i, --iterations <N>` - Number of benchmark iterations per measurement (default: 5)
+
 #### Import
 
 Exactly one of the following must be provided:

bin/miden-bench/src/benchmarks/mod.rs

@@ -1 +1,2 @@
+pub mod store;
 pub mod transaction;