writeonce/docs/examples/porch/middleware/idempotent.wo
shoney.arickathil d98ff82027 fix(porch-store): never replay a cached transient 5xx
- Miss path only marks a response a durable replay target (outcome 1)
  when status is 2xx/3xx; a 4xx/5xx gets outcome 3 instead
- Outcome 3's row is a one-shot relay: the scalar reply still can't carry
  a Resp (WO-E226), so the row exists only to hand the exact response
  back once, then idempotent.wo deletes it -- a retry with the same key
  is a genuine miss and re-executes, instead of caching a 500 for the
  24h default TTL
- Reviewer finding: caching any status meant a transient failure was
  replayed verbatim until TTL expiry, worse than no idempotency at all
- Gate leg 18d: FlakyHandler fails once then succeeds; same key twice
  must answer 500 then 200 -- confirmed failing (500, 500) before the
  fix, passing after

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
(cherry picked from commit e61015f2065a7c6aec6f5c2439e78b53f796bab3)
2026-09-15 01:15:30 +02:00

96 lines
4.5 KiB
Text

-- porch/middleware/idempotent.wo — idempotency, actor-run.
-- Iteration 1 of the porch track, porch-store task 4.
--
-- Rebuilt, not patched: the old before/after shape ran the handler and
-- stored the response afterward, so two simultaneous duplicates both
-- missed and both executed — before() has nothing to find until after()
-- runs, which is too late for the request that is racing it. The fix is
-- that the ACTOR runs the handler: Idempotent wraps the route's own
-- Handler and hands both the request and that handler to the pool. A
-- duplicate for the same key then simply waits in the actor's mailbox
-- (one message at a time) and is dequeued once the owner's receive has
-- already committed the row — no in-flight heuristic needed, because
-- there is no window where a duplicate can see "nothing yet".
--
-- `call`'s reply is a copyable scalar only (WO-E226): keypool.wo's kind-2
-- arm answers with the SAME pool_pack(count, remaining_ms) encoding kind-1
-- uses, never the response itself. The response travels through the
-- @table instead — the actor stores it, this file reads the same row
-- back and builds the Resp from it, so owner and duplicate answer with
-- byte-identical bytes structurally, not by careful bookkeeping.
use http
use json
-- Idempotent wraps a route's Handler. Registration: Idempotent { key_header:
-- "Idempotency-Key", pool: p, inner: CreateOrder {} } in place of the bare
-- handler in app.post(...). Only the digest allowlists content-type for
-- replay (never Set-Cookie, never Date) — cookies arrive in porch 2.
pub class Idempotent {
key_header: Text -- e.g., "Idempotency-Key"
pool: Pool
inner: Handler -- the real route handler; the actor runs this
include_body: Bool = true -- digest method+path+body, refuse a key reused with a different one
ttl: Int = 86_400_000_000 -- 24h in µs, lazy-expired on access
fn handle(req: Req) -> Resp {
let header_val = req.headers[self.key_header];
if header_val == nil { return self.inner.handle(req); }
let key = "idem:${self.key_header}:${header_val}";
let digest = "";
if self.include_body {
let digest_input = "${req.method}|${req.path}|${req.body}";
digest = base64_encode(bytes_slice(sha256(bytes_of_text(digest_input)), 0, 16));
}
let raw = try pool_begin(self.pool, key, digest, self.ttl, req, self.inner) catch (e) nil;
if raw == nil {
let r = Resp { status: 503, headers: {}, body: "{\"error\":\"idempotency store saturated\"}" };
set_header(r, "content-type", "application/json");
set_header(r, "retry-after", "1");
return r;
}
let outcome = raw / 1_000_000_000;
if outcome == 2 {
-- Same key, a different request: refuse rather than serve the
-- other request's response.
let r = Resp { status: 422, headers: {}, body: "{\"error\":\"idempotency key reused with a different request\"}" };
set_header(r, "content-type", "application/json");
return r;
}
-- outcome 1 (2xx/3xx, a durable replay target) or 3 (4xx/5xx, a
-- one-read relay only): either way the actor already committed this
-- row before returning, so it is there to read right now.
let hits = from k in IdempotencyKey where k.key == key take 1 select k;
if len(hits) == 0 { return server_error(); }
let row = hits[0];
let stored = json.decode(row.response) as IdempotentStoredResp;
if stored == nil { return server_error(); }
-- `.. ""` forces a fresh, independently-owned Text for every key/value
-- copied out of the decoded record: json.decode's Text values do not
-- survive being handed onward as-is once the decoded record itself
-- goes out of scope (a stale row read back corrupted mid-response
-- otherwise) — concat is documented to always allocate new owned text.
let hdrs: map<Text, Text> = {};
for k, v in stored.headers { hdrs[k .. ""] = v .. ""; }
let result = Resp { status: stored.status, headers: hdrs, body: stored.body .. "" };
if outcome == 3 {
-- Never a replay target: delete now so the next attempt with this
-- key (an immediate retry, most likely) is a genuine miss and
-- re-executes the handler, instead of caching a transient failure
-- for the TTL.
delete row;
}
return result;
}
}
-- JSON shape of IdempotencyKey.response. Allowlisted headers only:
-- content-type, never Set-Cookie or Date.
typedef IdempotentStoredResp = {
status: Int,
headers: map<Text, Text>,
body: Text
}