@hisoka-io/anon-rpc-worker 0.2.0 → 0.3.0

package/CHANGELOG.md

@@ -2,6 +2,30 @@
 
 All notable changes to `@hisoka-io/anon-rpc-worker`.
 
+## 0.3.0 (2026-10-06)
+
+Every 0.2.0 config stays valid. Works against entries on nox 0.4.0-rc.6 and on nodes with claim protocol v2.
+
+### Added
+
+- Config `hedgeAfterMs` (default 3,000, 0 turns it off): resendable reads send a second copy on another route at
+  about the observed p95 reply time; the first reply wins.
+- The client runs with the low-latency settings: binary, retaining, long-polling claims where the entry supports
+  them; concurrent claims with data blocks first; immediate resend on a lost reply; two transport-failure resends;
+  same-entry resends for a single bridge; a standby entry kept connected (claims share the primary connection).
+- Wallet calls first: the first registry check waits for the first call to settle (at most 15 s).
+- Compressed replies: upstream requests carry `accept-encoding: gzip` where the runtime has `DecompressionStream`,
+  and the worker inflates gzip bodies itself, capped at `maxResponseBytes`. A 414 KB `eth_getLogs` reply crosses the
+  mixnet as about 41 KB (2 reply packets instead of 14), a full block of 490 KB as about 81 KB (3 instead of 16).
+- Worker-local answers: `eth_chainId` and `net_version` per upstream URL once a second, independently routed
+  request agrees. Reads addressed by block hash (`eth_getBlockByHash`, `eth_getBlockReceipts`, `eth_getLogs`
+  with `blockHash`, EIP-1898 `{blockHash}` state reads, kept once two calls returned the same result) are behind
+  the `LocalAnswers` option `blockHashReads`, off in the worker until answers are checked against the hash. A hit answers at
+  once with the caller's JSON-RPC `id` and sends nothing. `Cache-Control: no-cache` or `no-store` skips it. An
+  empty log or receipt list is always fetched again, since an upstream behind the chain head answers `[]`.
+- The client's send-path settings: chunked packet writes, a send-window warm-up after each entry connection
+  opens, and concurrent calls spread over the pinned and standby entries.
+
 ## 0.2.0 (2026-10-06)
 
 Built on `@hisoka-io/nox-client` 0.6.0 (`kps.discovery`). Every 0.1.0 config stays valid.

package/README.md

@@ -41,6 +41,7 @@ optional, and every config valid for 0.1.0 stays valid:
 | `checkpoint` | | | reserved; any value fails with `bad-config` |
 | `logLevel` | `"debug"`, `"info"`, `"warn"`, `"error"` | `"info"` | |
 | `attemptTimeoutMs` | integer | 12,000 | 3,000-60,000 |
+| `hedgeAfterMs` | integer | 3,000 | 0-60,000. Resendable reads send a second copy on another route when no reply arrived after this long, raised to the observed p95 reply time and kept below `attemptTimeoutMs`; the first reply wins. 0 turns hedging off |
 | `callDeadlineMs` | integer | 25,000 | at least `attemptTimeoutMs`, at most 120,000 |
 | `maxConcurrentCalls` | integer | 16 | 1-64 |
 | `maxRequestBytes` | integer | 1,048,576 | 1,024-4,194,304 |
@@ -124,6 +125,22 @@ were read at, and when each member outside the snapshot was first seen. At most
 days. A cache that fails any check (another registry, too many records, one malformed record) is ignored whole, and
 a learned address is used only after the node's `/metadata.json` names the same member.
 
+## Release 0.2.0
+
+| | |
+|---|---|
+| `workerHash` (keccak-256 of `dist/anon-rpc-worker.js`) | `0x0a58f9915f686950072a4786249d396ecbf2194a39ac835effe8ea1a7c76f324` |
+| Size | 924,847 bytes; 268,945 bytes with `gzip -9 -n` |
+| Specifier | Ethereum Sepolia `0x29b4a6A8Cc11769531854d87f9F33EC63Efe8fe6` (`ImmutableWorkerSpecifier`, resolvers in [specifier/README.md](./specifier/README.md#deployments)) |
+| Snapshot | NoxRegistry at Arbitrum Sepolia block 316,207,920, keccak-256 `0xa351960e…94fbb2` |
+| nox-wasm module | 188,708 bytes, sha256 `5171416972ea4499fd4b4ea8bcadf002f4c10f10ed42cafb0d5a426121da9c4f` |
+| Built from | `nox-sdk` commit `ef019a716d7994820d7d63d6de3e60ae8f71e80d`, rustc 1.95.0, wasm-pack 0.13.1, wasm-bindgen 0.2.114, binaryen 117, esbuild 0.27.3 |
+
+Two builds in fresh containers of the pinned images (at two different paths) and a host build give these exact
+bytes (`scripts/verify-reproducible.sh --native`). Run it from the repository root as
+`packages/anon-rpc-worker/scripts/verify-reproducible.sh` (or `pnpm verify:reproducible` inside the package); host builds (`--native`,
+`--local`) use rustc 1.95.0, so set `RUSTUP_TOOLCHAIN=1.95.0` when the host default differs.
+
 ## What the bundle contains
 
 - `dist/anon-rpc-worker.js`: a single IIFE built by esbuild (target es2022, not minified, so it can be read