@reprova/sdk 0.5.0 → 0.6.0
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +75 -0
- package/dist/cjs/context.js +28 -0
- package/dist/cjs/distributionBaselinePoller.js +87 -0
- package/dist/cjs/distributionDrift.js +170 -0
- package/dist/cjs/drizzle.js +204 -0
- package/dist/cjs/frameworkAdapter.js +2 -0
- package/dist/cjs/httpFetch.js +59 -0
- package/dist/cjs/idempotency.js +132 -0
- package/dist/cjs/idempotencySettingsPoller.js +71 -0
- package/dist/cjs/index.js +96 -0
- package/dist/cjs/invariantInference.js +90 -0
- package/dist/cjs/invariantRules.js +142 -0
- package/dist/cjs/knex.js +86 -0
- package/dist/cjs/kysely.js +175 -0
- package/dist/cjs/legacyPrisma.js +130 -0
- package/dist/cjs/mikroorm.js +74 -0
- package/dist/cjs/mongoSchemaReporter.js +85 -0
- package/dist/cjs/mongoose.js +335 -0
- package/dist/cjs/nPlusOne.js +0 -0
- package/dist/cjs/nest.js +136 -0
- package/dist/cjs/nextjs.js +180 -0
- package/dist/cjs/package.json +1 -0
- package/dist/cjs/prisma.js +363 -0
- package/dist/cjs/proto.gen.js +4 -0
- package/dist/cjs/registry.js +100 -0
- package/dist/cjs/retryStorm.js +70 -0
- package/dist/cjs/rulesPoller.js +77 -0
- package/dist/cjs/sdk.js +1231 -0
- package/dist/cjs/sequelize.js +108 -0
- package/dist/cjs/shape.js +45 -0
- package/dist/cjs/stack.js +51 -0
- package/dist/cjs/transport.js +71 -0
- package/dist/cjs/typeorm.js +200 -0
- package/dist/context.d.ts +18 -0
- package/dist/context.d.ts.map +1 -1
- package/dist/context.js.map +1 -1
- package/dist/distributionBaselinePoller.d.ts +36 -0
- package/dist/distributionBaselinePoller.d.ts.map +1 -0
- package/dist/distributionBaselinePoller.js +84 -0
- package/dist/distributionBaselinePoller.js.map +1 -0
- package/dist/distributionDrift.d.ts +41 -0
- package/dist/distributionDrift.d.ts.map +1 -0
- package/dist/distributionDrift.js +161 -0
- package/dist/distributionDrift.js.map +1 -0
- package/dist/drizzle.d.ts.map +1 -1
- package/dist/drizzle.js +10 -4
- package/dist/drizzle.js.map +1 -1
- package/dist/frameworkAdapter.d.ts +5 -0
- package/dist/frameworkAdapter.d.ts.map +1 -0
- package/dist/frameworkAdapter.js +2 -0
- package/dist/frameworkAdapter.js.map +1 -0
- package/dist/httpFetch.d.ts +13 -0
- package/dist/httpFetch.d.ts.map +1 -0
- package/dist/httpFetch.js +54 -0
- package/dist/httpFetch.js.map +1 -0
- package/dist/idempotency.d.ts +26 -0
- package/dist/idempotency.d.ts.map +1 -0
- package/dist/idempotency.js +124 -0
- package/dist/idempotency.js.map +1 -0
- package/dist/idempotencySettingsPoller.d.ts +23 -0
- package/dist/idempotencySettingsPoller.d.ts.map +1 -0
- package/dist/idempotencySettingsPoller.js +68 -0
- package/dist/idempotencySettingsPoller.js.map +1 -0
- package/dist/index.d.ts +30 -5
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js +17 -3
- package/dist/index.js.map +1 -1
- package/dist/invariantInference.d.ts +28 -0
- package/dist/invariantInference.d.ts.map +1 -0
- package/dist/invariantInference.js +87 -0
- package/dist/invariantInference.js.map +1 -0
- package/dist/invariantRules.d.ts +16 -0
- package/dist/invariantRules.d.ts.map +1 -0
- package/dist/invariantRules.js +104 -0
- package/dist/invariantRules.js.map +1 -0
- package/dist/knex.d.ts.map +1 -1
- package/dist/knex.js +1 -0
- package/dist/knex.js.map +1 -1
- package/dist/kysely.d.ts.map +1 -1
- package/dist/kysely.js +2 -1
- package/dist/kysely.js.map +1 -1
- package/dist/legacyPrisma.d.ts +6 -0
- package/dist/legacyPrisma.d.ts.map +1 -0
- package/dist/legacyPrisma.js +128 -0
- package/dist/legacyPrisma.js.map +1 -0
- package/dist/mikroorm.d.ts +2 -0
- package/dist/mikroorm.d.ts.map +1 -0
- package/dist/mikroorm.js +72 -0
- package/dist/mikroorm.js.map +1 -0
- package/dist/mongoSchemaReporter.d.ts +23 -0
- package/dist/mongoSchemaReporter.d.ts.map +1 -0
- package/dist/mongoSchemaReporter.js +82 -0
- package/dist/mongoSchemaReporter.js.map +1 -0
- package/dist/mongoose.d.ts +48 -0
- package/dist/mongoose.d.ts.map +1 -0
- package/dist/mongoose.js +331 -0
- package/dist/mongoose.js.map +1 -0
- package/dist/nPlusOne.d.ts +52 -0
- package/dist/nPlusOne.d.ts.map +1 -0
- package/dist/nPlusOne.js +0 -0
- package/dist/nPlusOne.js.map +1 -0
- package/dist/nest.d.ts +13 -0
- package/dist/nest.d.ts.map +1 -1
- package/dist/nest.js +39 -0
- package/dist/nest.js.map +1 -1
- package/dist/nextjs.d.ts +27 -0
- package/dist/nextjs.d.ts.map +1 -0
- package/dist/nextjs.js +178 -0
- package/dist/nextjs.js.map +1 -0
- package/dist/prisma.d.ts +21 -5
- package/dist/prisma.d.ts.map +1 -1
- package/dist/prisma.js +247 -22
- package/dist/prisma.js.map +1 -1
- package/dist/proto.gen.d.ts +10 -2
- package/dist/proto.gen.d.ts.map +1 -1
- package/dist/registry.d.ts +1 -0
- package/dist/registry.d.ts.map +1 -1
- package/dist/registry.js +24 -0
- package/dist/registry.js.map +1 -1
- package/dist/retryStorm.d.ts +42 -0
- package/dist/retryStorm.d.ts.map +1 -0
- package/dist/retryStorm.js +65 -0
- package/dist/retryStorm.js.map +1 -0
- package/dist/rulesPoller.d.ts +26 -0
- package/dist/rulesPoller.d.ts.map +1 -0
- package/dist/rulesPoller.js +74 -0
- package/dist/rulesPoller.js.map +1 -0
- package/dist/sdk.d.ts +48 -7
- package/dist/sdk.d.ts.map +1 -1
- package/dist/sdk.js +619 -47
- package/dist/sdk.js.map +1 -1
- package/dist/sequelize.d.ts.map +1 -1
- package/dist/sequelize.js +4 -3
- package/dist/sequelize.js.map +1 -1
- package/dist/stack.d.ts.map +1 -1
- package/dist/stack.js +6 -3
- package/dist/stack.js.map +1 -1
- package/dist/transport.d.ts.map +1 -1
- package/dist/transport.js +2 -1
- package/dist/transport.js.map +1 -1
- package/dist/typeorm.d.ts +1 -0
- package/dist/typeorm.d.ts.map +1 -1
- package/dist/typeorm.js +133 -1
- package/dist/typeorm.js.map +1 -1
- package/package.json +14 -5
package/README.md
CHANGED
|
@@ -95,6 +95,39 @@ it ever reaches the captured bug — `repro run` reports `DIFFERENT FAILURE`
|
|
|
95
95
|
instead of reproducing the original error. `setAuthClaims` is a no-op
|
|
96
96
|
outside a Reprova request context, so it's safe to call unconditionally.
|
|
97
97
|
|
|
98
|
+
## Framework adapters (auth claims + migration_id for anything else)
|
|
99
|
+
|
|
100
|
+
`setAuthClaims` (above) and `sdk.setMigrationId` cover apps that can call
|
|
101
|
+
something from inside their own request/startup code. Some frameworks
|
|
102
|
+
resolve both through their own machinery instead — a DI container, a
|
|
103
|
+
non-Express request shape — with no single call site to hook. One call
|
|
104
|
+
covers both at once:
|
|
105
|
+
|
|
106
|
+
```ts
|
|
107
|
+
const sdk = Reprova.init({ dsn: process.env.REPROVA_DSN, release: gitSha });
|
|
108
|
+
|
|
109
|
+
sdk.registerFrameworkAdapter({
|
|
110
|
+
// Return undefined (not {}) to fall back to the built-in req.auth/req.user
|
|
111
|
+
// check instead of claiming "no auth" — useful if your framework only
|
|
112
|
+
// sometimes populates its own property.
|
|
113
|
+
extractAuthClaims: (rawRequest) => {
|
|
114
|
+
const req = rawRequest as { auth_context?: Record<string, unknown> };
|
|
115
|
+
return req.auth_context; // e.g. Medusa's own auth context
|
|
116
|
+
},
|
|
117
|
+
// Called once at registration — re-register (or call sdk.setMigrationId
|
|
118
|
+
// directly) if it can change during the process's lifetime.
|
|
119
|
+
resolveMigrationId: async () => {
|
|
120
|
+
const row = await db.query('SELECT name FROM my_migrations_table ORDER BY applied_at DESC LIMIT 1');
|
|
121
|
+
return row?.name ?? 'unknown';
|
|
122
|
+
},
|
|
123
|
+
});
|
|
124
|
+
```
|
|
125
|
+
|
|
126
|
+
Both fields are optional independently — supply only the one your framework
|
|
127
|
+
needs. Neither ever throws into request handling: `extractAuthClaims`
|
|
128
|
+
falls back to `req.auth ?? req.user` on error or `undefined`, and a
|
|
129
|
+
rejected `resolveMigrationId` just leaves `migration_id` as `'unknown'`.
|
|
130
|
+
|
|
98
131
|
## Prisma (optional enrichment — automatic footprints)
|
|
99
132
|
|
|
100
133
|
One call wires both automatic footprints and `migration_id`:
|
|
@@ -229,6 +262,48 @@ exported individually — pass the SAME logger instance to `createDrizzleCache`
|
|
|
229
262
|
to get the merge behavior; omit it to run the cache standalone (exact
|
|
230
263
|
SELECT PKs, no `where_shape`, no write coverage).
|
|
231
264
|
|
|
265
|
+
## N+1 query detection (automatic once any ORM adapter above is wired up)
|
|
266
|
+
|
|
267
|
+
Every adapter above feeds the same per-request footprint (one entry per
|
|
268
|
+
query call). The SDK watches that footprint on every response — success or
|
|
269
|
+
failure — for the same `(model, operation)` shape executed repeatedly as
|
|
270
|
+
separate calls in one request: a `for` loop firing one `findUnique` per
|
|
271
|
+
item instead of a single batched query. No exception, no 5xx — the request
|
|
272
|
+
usually returns 200 while doing far more DB round-trips than it should,
|
|
273
|
+
which is exactly the case nothing else here catches.
|
|
274
|
+
|
|
275
|
+
```ts
|
|
276
|
+
const sdk = Reprova.init({ dsn: process.env.REPROVA_DSN, release: gitSha, n1Threshold: 5 });
|
|
277
|
+
```
|
|
278
|
+
|
|
279
|
+
`n1Threshold` (default `5`) is how many times the same shape has to repeat
|
|
280
|
+
in one request before it's flagged; `0` disables detection entirely. It
|
|
281
|
+
can't mistake one efficient query for N+1: an `include`/join produces
|
|
282
|
+
exactly one footprint entry per relation regardless of how many rows it
|
|
283
|
+
returns, so only genuinely separate calls count.
|
|
284
|
+
|
|
285
|
+
## Retry storm detection (automatic — no wiring needed)
|
|
286
|
+
|
|
287
|
+
Every outbound `fetch()` call is already recorded per request (host,
|
|
288
|
+
method, path, status, outcome — used for outbound-call replay). The SDK
|
|
289
|
+
watches those calls, on every response, for the same `(host, method, path)`
|
|
290
|
+
called repeatedly with at least one failure among the attempts — a naive
|
|
291
|
+
retry loop hammering an already-struggling dependency instead of backing
|
|
292
|
+
off. Repetition alone isn't enough to flag: calling the same endpoint
|
|
293
|
+
several times and getting `200` every time is normal pagination or batch
|
|
294
|
+
fetching, not a storm — only repeats that include a failure (a `5xx`, or a
|
|
295
|
+
network error/timeout) count.
|
|
296
|
+
|
|
297
|
+
```ts
|
|
298
|
+
const sdk = Reprova.init({ dsn: process.env.REPROVA_DSN, release: gitSha, retryStormThreshold: 3 });
|
|
299
|
+
```
|
|
300
|
+
|
|
301
|
+
`retryStormThreshold` (default `3`, lower than `n1Threshold`'s `5` since
|
|
302
|
+
real backoff configs commonly cap around 3 attempts) is how many calls to
|
|
303
|
+
the same target it takes before a repeat-with-failures is flagged; `0`
|
|
304
|
+
disables it. The query string is stripped before grouping, since a real
|
|
305
|
+
retry commonly varies a request id or cache-busting param per attempt.
|
|
306
|
+
|
|
232
307
|
## NestJS (optional enrichment — original stacks + replay parity)
|
|
233
308
|
|
|
234
309
|
Nest's exception layer catches thrown errors before Express error middleware
|
|
@@ -0,0 +1,28 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.contextStorage = void 0;
|
|
4
|
+
exports.currentContext = currentContext;
|
|
5
|
+
exports.setAuthClaims = setAuthClaims;
|
|
6
|
+
const async_hooks_1 = require("async_hooks");
|
|
7
|
+
exports.contextStorage = new async_hooks_1.AsyncLocalStorage();
|
|
8
|
+
function currentContext() {
|
|
9
|
+
return exports.contextStorage.getStore();
|
|
10
|
+
}
|
|
11
|
+
/**
|
|
12
|
+
* Explicitly records the authenticated identity for the in-flight request.
|
|
13
|
+
* Auto-detection already covers `req.auth` (express-oauth2-jwt-bearer/Auth0)
|
|
14
|
+
* and `req.user` (Passport and most hand-rolled Express JWT middleware) —
|
|
15
|
+
* see detectAuthClaims in sdk.ts — so most apps need zero code changes. This
|
|
16
|
+
* is the escape hatch for anything else: a custom property, a non-Express
|
|
17
|
+
* framework. Call it from inside your own auth middleware once the token is
|
|
18
|
+
* verified, e.g. `setAuthClaims(decoded)` right after `jwt.verify`. Without
|
|
19
|
+
* captured claims one way or the other, replay has nothing to re-sign into a
|
|
20
|
+
* token and the app bounces the replayed request as unauthenticated before
|
|
21
|
+
* it ever reaches the captured bug. No-ops outside a Reprova request
|
|
22
|
+
* context.
|
|
23
|
+
*/
|
|
24
|
+
function setAuthClaims(claims) {
|
|
25
|
+
const ctx = currentContext();
|
|
26
|
+
if (ctx)
|
|
27
|
+
ctx.request.authClaims = claims;
|
|
28
|
+
}
|
|
@@ -0,0 +1,87 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.DistributionBaselinePoller = void 0;
|
|
4
|
+
const httpFetch_js_1 = require("./httpFetch.js");
|
|
5
|
+
const WARN_INTERVAL_MS = 60_000;
|
|
6
|
+
// Same start()/stop()/setInterval+unref shape as RulesPoller and
|
|
7
|
+
// IdempotencySettingsPoller — periodically fetches the tenant's current
|
|
8
|
+
// aggregate baselines (and the drift-detection on/off toggle) so a running
|
|
9
|
+
// process can check each response against a live baseline without a
|
|
10
|
+
// redeploy, and so drift detection stays off by default until an admin
|
|
11
|
+
// opts in from the dashboard. Falls back to disabled + no baselines before
|
|
12
|
+
// the first successful poll and on any failure, so a bad response or an
|
|
13
|
+
// unreachable control plane fails closed (no captures), not open (spurious
|
|
14
|
+
// captures from a stale or empty baseline).
|
|
15
|
+
class DistributionBaselinePoller {
|
|
16
|
+
enabled = false;
|
|
17
|
+
baselines = new Map();
|
|
18
|
+
timer = null;
|
|
19
|
+
lastWarnAt = 0;
|
|
20
|
+
endpoint;
|
|
21
|
+
apiKey;
|
|
22
|
+
intervalMs;
|
|
23
|
+
constructor(opts) {
|
|
24
|
+
this.endpoint = opts.endpoint;
|
|
25
|
+
this.apiKey = opts.apiKey;
|
|
26
|
+
this.intervalMs = opts.intervalMs ?? 60_000;
|
|
27
|
+
}
|
|
28
|
+
start() {
|
|
29
|
+
void this.refresh();
|
|
30
|
+
this.timer = setInterval(() => void this.refresh(), this.intervalMs);
|
|
31
|
+
if (this.timer.unref)
|
|
32
|
+
this.timer.unref();
|
|
33
|
+
}
|
|
34
|
+
stop() {
|
|
35
|
+
if (this.timer) {
|
|
36
|
+
clearInterval(this.timer);
|
|
37
|
+
this.timer = null;
|
|
38
|
+
}
|
|
39
|
+
}
|
|
40
|
+
isEnabled() {
|
|
41
|
+
return this.enabled;
|
|
42
|
+
}
|
|
43
|
+
getBaseline(route, field) {
|
|
44
|
+
return this.baselines.get(`${route}\x00${field}`);
|
|
45
|
+
}
|
|
46
|
+
async refresh() {
|
|
47
|
+
try {
|
|
48
|
+
const res = await (0, httpFetch_js_1.httpFetch)(this.endpoint, {
|
|
49
|
+
headers: { 'X-Reprova-Key': this.apiKey },
|
|
50
|
+
});
|
|
51
|
+
if (!res.ok) {
|
|
52
|
+
this.warnOnce(`reprova: distribution baselines fetch returned ${res.status}`);
|
|
53
|
+
return;
|
|
54
|
+
}
|
|
55
|
+
const body = (await res.json());
|
|
56
|
+
this.applyResponse(body);
|
|
57
|
+
}
|
|
58
|
+
catch (err) {
|
|
59
|
+
this.warnOnce(`reprova: distribution baselines fetch error: ${err.message}`);
|
|
60
|
+
}
|
|
61
|
+
}
|
|
62
|
+
// Applies an already-fetched response directly, with no network call —
|
|
63
|
+
// used by replay mode to seed a one-time snapshot the CLI fetched ahead of
|
|
64
|
+
// time (there's no live control plane to poll from inside a replayed
|
|
65
|
+
// app), so drift detection (checkDistributionDriftIfAny/
|
|
66
|
+
// captureWriteCountDriftIfAny in sdk.ts) sees exactly the same shape it
|
|
67
|
+
// would from a real refresh().
|
|
68
|
+
seed(body) {
|
|
69
|
+
this.applyResponse(body);
|
|
70
|
+
}
|
|
71
|
+
applyResponse(body) {
|
|
72
|
+
this.enabled = body.enabled ?? false;
|
|
73
|
+
const next = new Map();
|
|
74
|
+
for (const b of body.baselines ?? []) {
|
|
75
|
+
next.set(`${b.route}\x00${b.field}`, { sampleCount: b.sample_count, mean: b.mean, stddev: b.stddev });
|
|
76
|
+
}
|
|
77
|
+
this.baselines = next;
|
|
78
|
+
}
|
|
79
|
+
warnOnce(msg) {
|
|
80
|
+
const now = Date.now();
|
|
81
|
+
if (now - this.lastWarnAt > WARN_INTERVAL_MS) {
|
|
82
|
+
this.lastWarnAt = now;
|
|
83
|
+
console.warn(msg);
|
|
84
|
+
}
|
|
85
|
+
}
|
|
86
|
+
}
|
|
87
|
+
exports.DistributionBaselinePoller = DistributionBaselinePoller;
|
|
@@ -0,0 +1,170 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
// Distributional drift, numeric-only v1 slice: baseline the typical range
|
|
3
|
+
// of a numeric response field per (route, field) using Welford's online
|
|
4
|
+
// mean/variance algorithm, and flag a value that's implausibly far from
|
|
5
|
+
// that baseline — a `total` field that's normally 10-500 suddenly
|
|
6
|
+
// returning -40, with no rule authored and no equality asserted anywhere.
|
|
7
|
+
// Unlike business-invariant-rules and idempotency-key matching, this is a
|
|
8
|
+
// statistical guess with real false-positive risk (there's no
|
|
9
|
+
// "asymptotically safe" case the way an app-asserted equality is), so it's
|
|
10
|
+
// off by default and the thresholds below are fixed, not yet
|
|
11
|
+
// tenant-configurable.
|
|
12
|
+
//
|
|
13
|
+
// Deliberately scoped to numeric fields only for this slice — categorical
|
|
14
|
+
// novelty (a status enum emitting a never-seen value) is a structurally
|
|
15
|
+
// different mechanism (tracking a bounded set of distinct values, not a
|
|
16
|
+
// running mean/variance) and is left for a later slice.
|
|
17
|
+
//
|
|
18
|
+
// Observes the top-level fields of a JSON object response body only — not
|
|
19
|
+
// nested objects/arrays, and not array-shaped top-level responses (a list
|
|
20
|
+
// endpoint's items each have their own values; there's no single baseline
|
|
21
|
+
// for "the array"). That keeps the walk trivial and the key space bounded
|
|
22
|
+
// to (route, field) instead of an unbounded set of JSON paths.
|
|
23
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
24
|
+
exports.DistributionDriftReporter = exports.WriteCountDrift = exports.ValueDistributionDrift = void 0;
|
|
25
|
+
exports.newWelfordState = newWelfordState;
|
|
26
|
+
exports.updateWelfordState = updateWelfordState;
|
|
27
|
+
exports.isDrift = isDrift;
|
|
28
|
+
exports.extractTopLevelNumericFields = extractTopLevelNumericFields;
|
|
29
|
+
const httpFetch_js_1 = require("./httpFetch.js");
|
|
30
|
+
const MIN_SAMPLES = 200;
|
|
31
|
+
const SIGMA_MULTIPLIER = 6;
|
|
32
|
+
class ValueDistributionDrift extends Error {
|
|
33
|
+
}
|
|
34
|
+
exports.ValueDistributionDrift = ValueDistributionDrift;
|
|
35
|
+
// Same mean/variance mechanism, same MIN_SAMPLES/SIGMA_MULTIPLIER thresholds,
|
|
36
|
+
// entirely different observation site: prisma.ts's observeWriteCountDrift
|
|
37
|
+
// baselines a Prisma updateMany/deleteMany/createMany's affected-row count
|
|
38
|
+
// per (route, operation:model) instead of an HTTP response field's value. Shares
|
|
39
|
+
// WelfordState/updateWelfordState/isDrift directly — the math doesn't care
|
|
40
|
+
// what number it's baselining — and DistributionDriftReporter/
|
|
41
|
+
// DistributionBaselinePoller are already fully generic (record/getBaseline
|
|
42
|
+
// take arbitrary string keys), so a second instance of each, pointed at a
|
|
43
|
+
// separate control-plane endpoint, is all that's needed there too. Kept as
|
|
44
|
+
// a distinct Error class/trigger since it's a conceptually separate
|
|
45
|
+
// question ("did this write touch a weird number of rows" vs. "is this
|
|
46
|
+
// response value weird") with its own independent dashboard toggle.
|
|
47
|
+
class WriteCountDrift extends Error {
|
|
48
|
+
}
|
|
49
|
+
exports.WriteCountDrift = WriteCountDrift;
|
|
50
|
+
function newWelfordState() {
|
|
51
|
+
return { count: 0, mean: 0, m2: 0, min: Infinity, max: -Infinity };
|
|
52
|
+
}
|
|
53
|
+
// Standard single-pass Welford update — mean and M2 (sum of squared
|
|
54
|
+
// deviations from the running mean) stay numerically stable across an
|
|
55
|
+
// unbounded stream, unlike naively accumulating sum/sum-of-squares.
|
|
56
|
+
function updateWelfordState(state, x) {
|
|
57
|
+
state.count++;
|
|
58
|
+
const delta = x - state.mean;
|
|
59
|
+
state.mean += delta / state.count;
|
|
60
|
+
const delta2 = x - state.mean;
|
|
61
|
+
state.m2 += delta * delta2;
|
|
62
|
+
if (x < state.min)
|
|
63
|
+
state.min = x;
|
|
64
|
+
if (x > state.max)
|
|
65
|
+
state.max = x;
|
|
66
|
+
}
|
|
67
|
+
// A value is drift when the baseline has enough samples to trust AND the
|
|
68
|
+
// value is more than SIGMA_MULTIPLIER standard deviations from the mean.
|
|
69
|
+
// A zero-variance baseline (every sample identical so far) treats ANY
|
|
70
|
+
// different value as infinite deviation — flagged, since after MIN_SAMPLES
|
|
71
|
+
// consistent observations a field that never varied suddenly doing so is
|
|
72
|
+
// exactly the kind of thing worth surfacing, not a numerical edge case to
|
|
73
|
+
// special-case away.
|
|
74
|
+
function isDrift(value, baseline) {
|
|
75
|
+
if (baseline.sampleCount < MIN_SAMPLES)
|
|
76
|
+
return false;
|
|
77
|
+
const deviation = baseline.stddev === 0 ? (value !== baseline.mean ? Infinity : 0) : Math.abs(value - baseline.mean) / baseline.stddev;
|
|
78
|
+
return deviation > SIGMA_MULTIPLIER;
|
|
79
|
+
}
|
|
80
|
+
// Extracts finite numeric values from the response body's OWN top-level
|
|
81
|
+
// keys — null/NaN/Infinity are skipped (not a "value", nothing to
|
|
82
|
+
// baseline), and a non-object or array body yields nothing (see module doc
|
|
83
|
+
// for why array-shaped top-level responses are out of scope for this slice).
|
|
84
|
+
function extractTopLevelNumericFields(body) {
|
|
85
|
+
if (!body || typeof body !== 'object' || Array.isArray(body))
|
|
86
|
+
return {};
|
|
87
|
+
const out = {};
|
|
88
|
+
for (const [k, v] of Object.entries(body)) {
|
|
89
|
+
if (typeof v === 'number' && Number.isFinite(v))
|
|
90
|
+
out[k] = v;
|
|
91
|
+
}
|
|
92
|
+
return out;
|
|
93
|
+
}
|
|
94
|
+
const WARN_INTERVAL_MS = 60_000;
|
|
95
|
+
// Accumulates local Welford state per (route, field) since the last flush
|
|
96
|
+
// and periodically reports it as a delta — same start/stop/periodic-flush
|
|
97
|
+
// shape as BatchTransport/InferenceReporter. The control plane merges each
|
|
98
|
+
// report into the tenant's cumulative baseline via the parallel-variance
|
|
99
|
+
// formula; a dropped/failed flush only ever loses that batch's
|
|
100
|
+
// contribution (undercounts), never double-reports on retry, since there
|
|
101
|
+
// is no retry — same trade-off every other reporter in this SDK already
|
|
102
|
+
// makes.
|
|
103
|
+
class DistributionDriftReporter {
|
|
104
|
+
states = new Map();
|
|
105
|
+
timer = null;
|
|
106
|
+
lastWarnAt = 0;
|
|
107
|
+
endpoint;
|
|
108
|
+
apiKey;
|
|
109
|
+
flushIntervalMs;
|
|
110
|
+
constructor(opts) {
|
|
111
|
+
this.endpoint = opts.endpoint;
|
|
112
|
+
this.apiKey = opts.apiKey;
|
|
113
|
+
this.flushIntervalMs = opts.flushIntervalMs ?? 30_000;
|
|
114
|
+
}
|
|
115
|
+
start() {
|
|
116
|
+
this.timer = setInterval(() => void this.flush(), this.flushIntervalMs);
|
|
117
|
+
if (this.timer.unref)
|
|
118
|
+
this.timer.unref();
|
|
119
|
+
}
|
|
120
|
+
stop() {
|
|
121
|
+
if (this.timer) {
|
|
122
|
+
clearInterval(this.timer);
|
|
123
|
+
this.timer = null;
|
|
124
|
+
}
|
|
125
|
+
}
|
|
126
|
+
record(route, field, value) {
|
|
127
|
+
const key = `${route}\x00${field}`;
|
|
128
|
+
let entry = this.states.get(key);
|
|
129
|
+
if (!entry) {
|
|
130
|
+
entry = { route, field, state: newWelfordState() };
|
|
131
|
+
this.states.set(key, entry);
|
|
132
|
+
}
|
|
133
|
+
updateWelfordState(entry.state, value);
|
|
134
|
+
}
|
|
135
|
+
async flush() {
|
|
136
|
+
if (this.states.size === 0)
|
|
137
|
+
return;
|
|
138
|
+
const observations = [...this.states.values()].map(({ route, field, state }) => ({
|
|
139
|
+
route,
|
|
140
|
+
field,
|
|
141
|
+
count: state.count,
|
|
142
|
+
mean: state.mean,
|
|
143
|
+
m2: state.m2,
|
|
144
|
+
min: state.min,
|
|
145
|
+
max: state.max,
|
|
146
|
+
}));
|
|
147
|
+
this.states.clear();
|
|
148
|
+
try {
|
|
149
|
+
const res = await (0, httpFetch_js_1.httpFetch)(this.endpoint, {
|
|
150
|
+
method: 'POST',
|
|
151
|
+
headers: { 'Content-Type': 'application/json', 'X-Reprova-Key': this.apiKey },
|
|
152
|
+
body: JSON.stringify({ observations }),
|
|
153
|
+
});
|
|
154
|
+
if (!res.ok) {
|
|
155
|
+
this.warnOnce(`reprova: distribution baseline report returned ${res.status}`);
|
|
156
|
+
}
|
|
157
|
+
}
|
|
158
|
+
catch (err) {
|
|
159
|
+
this.warnOnce(`reprova: distribution baseline report error: ${err.message}`);
|
|
160
|
+
}
|
|
161
|
+
}
|
|
162
|
+
warnOnce(msg) {
|
|
163
|
+
const now = Date.now();
|
|
164
|
+
if (now - this.lastWarnAt > WARN_INTERVAL_MS) {
|
|
165
|
+
this.lastWarnAt = now;
|
|
166
|
+
console.warn(msg);
|
|
167
|
+
}
|
|
168
|
+
}
|
|
169
|
+
}
|
|
170
|
+
exports.DistributionDriftReporter = DistributionDriftReporter;
|
|
@@ -0,0 +1,204 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
3
|
+
exports.createDrizzleLogger = createDrizzleLogger;
|
|
4
|
+
exports.createDrizzleCache = createDrizzleCache;
|
|
5
|
+
exports.createDrizzleFootprint = createDrizzleFootprint;
|
|
6
|
+
const node_crypto_1 = require("node:crypto");
|
|
7
|
+
const context_js_1 = require("./context.js");
|
|
8
|
+
const OP_BY_KEYWORD = {
|
|
9
|
+
select: 'select',
|
|
10
|
+
insert: 'insert',
|
|
11
|
+
update: 'update',
|
|
12
|
+
delete: 'delete',
|
|
13
|
+
};
|
|
14
|
+
// First SQL keyword → operation.
|
|
15
|
+
function opOf(sql) {
|
|
16
|
+
const m = /^\s*(select|insert|update|delete)\b/i.exec(sql);
|
|
17
|
+
return m ? OP_BY_KEYWORD[m[1].toLowerCase()] : undefined;
|
|
18
|
+
}
|
|
19
|
+
// The principal table, from the op-appropriate clause.
|
|
20
|
+
function tableOf(sql, op) {
|
|
21
|
+
let m = null;
|
|
22
|
+
if (op === 'select' || op === 'delete')
|
|
23
|
+
m = /\bfrom\s+"?([A-Za-z0-9_]+)"?/i.exec(sql);
|
|
24
|
+
else if (op === 'insert')
|
|
25
|
+
m = /\binto\s+"?([A-Za-z0-9_]+)"?/i.exec(sql);
|
|
26
|
+
else if (op === 'update')
|
|
27
|
+
m = /\bupdate\s+"?([A-Za-z0-9_]+)"?/i.exec(sql);
|
|
28
|
+
return m ? m[1] : undefined;
|
|
29
|
+
}
|
|
30
|
+
// Column names referenced in the WHERE clause — the right-hand identifier of
|
|
31
|
+
// each `"table"."col"` (or a bare `"col"`). Names only; values are `?`.
|
|
32
|
+
function whereColumns(sql) {
|
|
33
|
+
const wIdx = sql.search(/\bwhere\b/i);
|
|
34
|
+
if (wIdx < 0)
|
|
35
|
+
return [];
|
|
36
|
+
const tail = sql.slice(wIdx + 5);
|
|
37
|
+
const cols = new Set();
|
|
38
|
+
const re = /"([A-Za-z0-9_]+)"(?:\s*\.\s*"([A-Za-z0-9_]+)")?/g;
|
|
39
|
+
let m;
|
|
40
|
+
while ((m = re.exec(tail)) !== null)
|
|
41
|
+
cols.add(m[2] ?? m[1]);
|
|
42
|
+
return [...cols].sort();
|
|
43
|
+
}
|
|
44
|
+
function extractPks(rows, pkCols) {
|
|
45
|
+
if (pkCols.length === 0)
|
|
46
|
+
return [];
|
|
47
|
+
const seen = new Set();
|
|
48
|
+
const pks = [];
|
|
49
|
+
for (const row of rows) {
|
|
50
|
+
const vals = pkCols.map((c) => row[c]);
|
|
51
|
+
if (vals.some((v) => v === undefined))
|
|
52
|
+
continue; // column not selected — can't form a PK
|
|
53
|
+
const pk = vals.map((v) => String(v)).join(':');
|
|
54
|
+
if (!seen.has(pk)) {
|
|
55
|
+
seen.add(pk);
|
|
56
|
+
pks.push(pk);
|
|
57
|
+
}
|
|
58
|
+
}
|
|
59
|
+
return pks;
|
|
60
|
+
}
|
|
61
|
+
// Drizzle's own cache key: SHA-256(sql + '-' + JSON.stringify(params)) — see
|
|
62
|
+
// drizzle-orm/cache/core/cache.js. Reimplemented (not imported) so a query's
|
|
63
|
+
// hash computed here from `logger.logQuery`'s (sql, params) matches the hash
|
|
64
|
+
// Drizzle later passes to `cache.put()` for the SAME query, letting the two
|
|
65
|
+
// hooks correlate without ever importing 'drizzle-orm'.
|
|
66
|
+
// Node's own crypto module (not WebCrypto's `crypto.subtle`, which only
|
|
67
|
+
// became a stable global in Node 19+) — createHash has existed since
|
|
68
|
+
// Node's earliest crypto module, so this works on any Node version this
|
|
69
|
+
// SDK targets. Stays async (a plain value from an async function resolves
|
|
70
|
+
// immediately) so callers don't need to change.
|
|
71
|
+
async function hashQuery(sql, params) {
|
|
72
|
+
const data = `${sql}-${JSON.stringify(params)}`;
|
|
73
|
+
return (0, node_crypto_1.createHash)('sha256').update(data).digest('hex');
|
|
74
|
+
}
|
|
75
|
+
// A pending SELECT footprint entry, keyed by query hash, waiting to be
|
|
76
|
+
// upgraded by a paired cache's put(). Linked to its logger via WeakMap so the
|
|
77
|
+
// map is GC'd along with the logger if the app never pairs a cache to it.
|
|
78
|
+
const pendingByLogger = new WeakMap();
|
|
79
|
+
// Creates a Drizzle-compatible logger that records footprints.
|
|
80
|
+
// Usage: drizzle(client, { logger: createDrizzleLogger() })
|
|
81
|
+
function createDrizzleLogger(opts = {}) {
|
|
82
|
+
const captureWhereShape = opts.captureWhereShape !== false;
|
|
83
|
+
const forward = opts.forward;
|
|
84
|
+
const pending = new Map();
|
|
85
|
+
const logger = {
|
|
86
|
+
logQuery(query, params) {
|
|
87
|
+
try {
|
|
88
|
+
const ctx = context_js_1.contextStorage.getStore();
|
|
89
|
+
if (ctx) {
|
|
90
|
+
const op = opOf(query);
|
|
91
|
+
const table = op ? tableOf(query, op) : undefined;
|
|
92
|
+
if (op && table) {
|
|
93
|
+
const where = captureWhereShape ? whereColumns(query).join(',') || undefined : undefined;
|
|
94
|
+
const entry = {
|
|
95
|
+
model: table,
|
|
96
|
+
op,
|
|
97
|
+
pks: [],
|
|
98
|
+
where_shape: where,
|
|
99
|
+
count: 0,
|
|
100
|
+
note: 'drizzle: shape-only (no result access — PKs/count unavailable)',
|
|
101
|
+
resolved: true,
|
|
102
|
+
};
|
|
103
|
+
ctx.footprint.push(entry);
|
|
104
|
+
// A paired cache can upgrade this SAME entry once the real rows
|
|
105
|
+
// are known — see createDrizzleCache(). Mutations get nothing
|
|
106
|
+
// more from Drizzle's cache hook, so they stay as pushed above.
|
|
107
|
+
if (op === 'select') {
|
|
108
|
+
hashQuery(query, params)
|
|
109
|
+
.then((h) => pending.set(h, entry))
|
|
110
|
+
.catch(() => { });
|
|
111
|
+
}
|
|
112
|
+
}
|
|
113
|
+
}
|
|
114
|
+
}
|
|
115
|
+
catch { /* never break a query */ }
|
|
116
|
+
try {
|
|
117
|
+
forward?.logQuery(query, params);
|
|
118
|
+
}
|
|
119
|
+
catch { /* ignore */ }
|
|
120
|
+
},
|
|
121
|
+
};
|
|
122
|
+
pendingByLogger.set(logger, pending);
|
|
123
|
+
return logger;
|
|
124
|
+
}
|
|
125
|
+
// node-postgres (`pg`) QueryResult vs. mysql2's `[rows, fields]` tuple — the
|
|
126
|
+
// two response shapes `cache.put()` actually sees for Reprova's two real
|
|
127
|
+
// customer-database dialects (Postgres, MySQL; see agent/internal/dialect).
|
|
128
|
+
// Unrecognized shapes degrade to "no rows found" rather than guessing.
|
|
129
|
+
function rowsFromResponse(response) {
|
|
130
|
+
if (Array.isArray(response)) {
|
|
131
|
+
const [rows] = response; // mysql2: client.query() resolves [rows, fields]
|
|
132
|
+
return Array.isArray(rows) ? rows : undefined;
|
|
133
|
+
}
|
|
134
|
+
const rows = response?.rows;
|
|
135
|
+
return Array.isArray(rows) ? rows : undefined;
|
|
136
|
+
}
|
|
137
|
+
// Creates a Drizzle Cache that observes (never serves) query results to
|
|
138
|
+
// upgrade SELECT footprints with exact primary keys and row counts.
|
|
139
|
+
// Usage: drizzle(client, { cache: createDrizzleCache({ logger }) })
|
|
140
|
+
function createDrizzleCache(opts = {}) {
|
|
141
|
+
const primaryKey = opts.primaryKey ?? (() => ['id']);
|
|
142
|
+
const pending = opts.logger ? pendingByLogger.get(opts.logger) : undefined;
|
|
143
|
+
return {
|
|
144
|
+
// 'all' — every SELECT is observed automatically; apps never need to
|
|
145
|
+
// call Drizzle's own `.$withCache()` for this to work.
|
|
146
|
+
strategy() {
|
|
147
|
+
return 'all';
|
|
148
|
+
},
|
|
149
|
+
// Always a miss: this cache exists to OBSERVE results, never to REPLACE a
|
|
150
|
+
// real query with a stale one.
|
|
151
|
+
async get() {
|
|
152
|
+
return undefined;
|
|
153
|
+
},
|
|
154
|
+
async put(hashedQuery, response, tables) {
|
|
155
|
+
try {
|
|
156
|
+
if (!tables || tables.length === 0)
|
|
157
|
+
return;
|
|
158
|
+
const rows = rowsFromResponse(response);
|
|
159
|
+
if (!rows)
|
|
160
|
+
return;
|
|
161
|
+
const primary = tables[0];
|
|
162
|
+
const pks = extractPks(rows, primaryKey(primary));
|
|
163
|
+
const count = rows.length;
|
|
164
|
+
const note = count === 0 ? 'returned 0 rows' : undefined;
|
|
165
|
+
const linked = pending?.get(hashedQuery);
|
|
166
|
+
const ctx = context_js_1.contextStorage.getStore();
|
|
167
|
+
if (linked) {
|
|
168
|
+
pending.delete(hashedQuery);
|
|
169
|
+
linked.model = primary;
|
|
170
|
+
linked.pks = pks;
|
|
171
|
+
linked.count = count;
|
|
172
|
+
linked.note = note;
|
|
173
|
+
}
|
|
174
|
+
else if (ctx) {
|
|
175
|
+
ctx.footprint.push({ model: primary, op: 'select', pks, count, note, resolved: true });
|
|
176
|
+
}
|
|
177
|
+
// Joined/subquery tables: recorded as referenced, without PKs — same
|
|
178
|
+
// convention as the Kysely adapter (result columns can't be reliably
|
|
179
|
+
// attributed to a joined table's own primary key).
|
|
180
|
+
if (ctx) {
|
|
181
|
+
for (const t of tables.slice(1)) {
|
|
182
|
+
ctx.footprint.push({ model: t, op: 'select', pks: [], count: 0, note: 'referenced (join/subquery)', resolved: true });
|
|
183
|
+
}
|
|
184
|
+
}
|
|
185
|
+
}
|
|
186
|
+
catch { /* never break a query */ }
|
|
187
|
+
},
|
|
188
|
+
// Drizzle's mutation hook gets table names only — no rows, no count, and
|
|
189
|
+
// it races the real query (Promise.all internally). Nothing usable to
|
|
190
|
+
// record here; mutations are already captured (shape-only) by the paired
|
|
191
|
+
// logger.
|
|
192
|
+
async onMutate() { },
|
|
193
|
+
};
|
|
194
|
+
}
|
|
195
|
+
// Convenience: builds a linked logger+cache pair in one call, the
|
|
196
|
+
// recommended way to get exact SELECT pks/count without double-counting.
|
|
197
|
+
// Usage:
|
|
198
|
+
// const { logger, cache } = createDrizzleFootprint();
|
|
199
|
+
// const db = drizzle(client, { logger, cache });
|
|
200
|
+
function createDrizzleFootprint(opts = {}) {
|
|
201
|
+
const logger = createDrizzleLogger({ captureWhereShape: opts.captureWhereShape });
|
|
202
|
+
const cache = createDrizzleCache({ primaryKey: opts.primaryKey, logger });
|
|
203
|
+
return { logger, cache };
|
|
204
|
+
}
|
|
@@ -0,0 +1,59 @@
|
|
|
1
|
+
"use strict";
|
|
2
|
+
var __importDefault = (this && this.__importDefault) || function (mod) {
|
|
3
|
+
return (mod && mod.__esModule) ? mod : { "default": mod };
|
|
4
|
+
};
|
|
5
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
|
6
|
+
exports.httpFetch = httpFetch;
|
|
7
|
+
const node_http_1 = __importDefault(require("node:http"));
|
|
8
|
+
const node_https_1 = __importDefault(require("node:https"));
|
|
9
|
+
// Node 18+ ships a global fetch; older Node (found live: Node 14, still a
|
|
10
|
+
// real LTS for legacy apps) has none at all. node-fetch isn't a safe
|
|
11
|
+
// replacement here — v3+ is ESM-only, the exact same require(ESM) trap
|
|
12
|
+
// @marcbachmann/cel-js caused (see invariantRules.ts's own doc comment on
|
|
13
|
+
// loadEnv). Every internal caller in this SDK only ever needs a small,
|
|
14
|
+
// stable subset of the Fetch API (GET/POST, headers, a string body,
|
|
15
|
+
// .ok/.status/.json()/.text()), so this implements just that on top of
|
|
16
|
+
// Node's own always-available http/https modules, and delegates to a real
|
|
17
|
+
// global fetch when one exists (the common case — zero overhead, zero
|
|
18
|
+
// behavior change on any Node version that already has one).
|
|
19
|
+
function httpFetch(url, init) {
|
|
20
|
+
const globalFetch = globalThis.fetch;
|
|
21
|
+
if (typeof globalFetch === 'function') {
|
|
22
|
+
return globalFetch(url, init);
|
|
23
|
+
}
|
|
24
|
+
return new Promise((resolve, reject) => {
|
|
25
|
+
let parsed;
|
|
26
|
+
try {
|
|
27
|
+
parsed = new URL(url);
|
|
28
|
+
}
|
|
29
|
+
catch (err) {
|
|
30
|
+
reject(err);
|
|
31
|
+
return;
|
|
32
|
+
}
|
|
33
|
+
const lib = parsed.protocol === 'https:' ? node_https_1.default : node_http_1.default;
|
|
34
|
+
const req = lib.request({
|
|
35
|
+
hostname: parsed.hostname,
|
|
36
|
+
port: parsed.port || undefined,
|
|
37
|
+
path: `${parsed.pathname}${parsed.search}`,
|
|
38
|
+
method: init?.method ?? 'GET',
|
|
39
|
+
headers: init?.headers,
|
|
40
|
+
}, (res) => {
|
|
41
|
+
const chunks = [];
|
|
42
|
+
res.on('data', (chunk) => chunks.push(chunk));
|
|
43
|
+
res.on('end', () => {
|
|
44
|
+
const text = Buffer.concat(chunks).toString('utf8');
|
|
45
|
+
const status = res.statusCode ?? 0;
|
|
46
|
+
resolve({
|
|
47
|
+
ok: status >= 200 && status < 300,
|
|
48
|
+
status,
|
|
49
|
+
json: async () => JSON.parse(text),
|
|
50
|
+
text: async () => text,
|
|
51
|
+
});
|
|
52
|
+
});
|
|
53
|
+
});
|
|
54
|
+
req.on('error', reject);
|
|
55
|
+
if (init?.body)
|
|
56
|
+
req.write(init.body);
|
|
57
|
+
req.end();
|
|
58
|
+
});
|
|
59
|
+
}
|