@genroc/eval-node 0.0.0-edge.86dae06 → 0.0.0-edge.a5e8589
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 +39 -32
- package/dist/eval.js +4 -3
- package/dist/import.js +12 -16
- package/dist/realm.js +66 -75
- package/dist/worker.js +6 -1
- package/eval.ts +5 -3
- package/import.ts +12 -16
- package/package.json +1 -1
- package/realm.ts +61 -77
- package/worker.ts +7 -1
package/README.md
CHANGED
|
@@ -96,15 +96,18 @@ The `input` of the external task IS the evaluation request:
|
|
|
96
96
|
|
|
97
97
|
```jsonc
|
|
98
98
|
{
|
|
99
|
-
"code": "
|
|
100
|
-
"input": { "amount": 250 },
|
|
101
|
-
"timeout_ms": 5000
|
|
99
|
+
"code": "export default (input) => ({ fee: input.amount * 0.1 });", // required — a module
|
|
100
|
+
"input": { "amount": 250 }, // optional — its argument
|
|
101
|
+
"timeout_ms": 5000 // optional — default 5000
|
|
102
102
|
}
|
|
103
103
|
```
|
|
104
104
|
|
|
105
|
-
`code` is
|
|
106
|
-
|
|
107
|
-
|
|
105
|
+
`code` is an **ES module**, and the evaluator imports it and calls its **default export** with
|
|
106
|
+
`input`. The value it returns reaches genroc; the function may be async, and so may the module's
|
|
107
|
+
top level. A module exporting anything else — or nothing — is a `compile_error`: there is
|
|
108
|
+
nothing to call, and so is one that will not parse; a throw from its top level is a `threw`
|
|
109
|
+
like any other. `import` of a node builtin resolves as it does anywhere else, and that is what
|
|
110
|
+
a bundled builtin lands on.
|
|
108
111
|
|
|
109
112
|
## The answer — the failure kind IS the error code
|
|
110
113
|
|
|
@@ -140,12 +143,14 @@ no retry policy to write.
|
|
|
140
143
|
type: external
|
|
141
144
|
input:
|
|
142
145
|
code: |
|
|
143
|
-
|
|
144
|
-
|
|
145
|
-
|
|
146
|
-
|
|
146
|
+
export default function (input) {
|
|
147
|
+
if (input.amount > 100) {
|
|
148
|
+
const e = new Error('amount over the limit');
|
|
149
|
+
e.name = 'LimitExceeded';
|
|
150
|
+
throw e;
|
|
151
|
+
}
|
|
152
|
+
return { fee: input.amount * 0.1 };
|
|
147
153
|
}
|
|
148
|
-
return { fee: input.amount * 0.1 };
|
|
149
154
|
input: "$: input"
|
|
150
155
|
result_schema: { type: object, properties: { fee: { type: number } }, required: [fee] }
|
|
151
156
|
raises:
|
|
@@ -227,12 +232,14 @@ the inferred type of what the definition passes; `Output` is what it declares
|
|
|
227
232
|
`tsc --noEmit`, and bundles — so **a type error is a failed apply**, and a stored definition
|
|
228
233
|
cannot hold code that failed to typecheck.
|
|
229
234
|
|
|
230
|
-
The bundle is
|
|
231
|
-
|
|
232
|
-
|
|
233
|
-
|
|
234
|
-
|
|
235
|
-
|
|
235
|
+
The bundle is one self-contained ES module whose default export is the author's own — nothing
|
|
236
|
+
wraps or rewrites it, so what the editor checks and what the realm runs are the same module, and
|
|
237
|
+
**a missing default export is a failed apply** rather than a fault at run time. Imports resolve
|
|
238
|
+
through TypeScript under the same config the check ran with, so a `paths` alias that typechecks
|
|
239
|
+
also bundles. They are inlined at build time, so the string a definition version stores is
|
|
240
|
+
self-contained forever — with one exception: **node builtins stay as imports**, which the realm
|
|
241
|
+
resolves. A package is frozen into the definition; `node:fs` is resolved by whatever runner
|
|
242
|
+
executes it.
|
|
236
243
|
|
|
237
244
|
### Your tsconfig, your types
|
|
238
245
|
|
|
@@ -258,8 +265,9 @@ never read by genroc, because genctl doubles every `$` on splice.
|
|
|
258
265
|
|
|
259
266
|
## The realm — one Worker per execution
|
|
260
267
|
|
|
261
|
-
`evaluate()` starts a Worker (`realm.ts`), posts the code into it
|
|
262
|
-
`
|
|
268
|
+
`evaluate()` starts a Worker (`realm.ts`), posts the code into it — where a module loader hook
|
|
269
|
+
serves it to `import()`, so the engine numbers the script's frames from the author's own source —
|
|
270
|
+
and races the reply against the budget; `terminate()` runs on every path. That thread is what the contract rests on, and it buys
|
|
263
271
|
exactly three things the previous in-process evaluator could not:
|
|
264
272
|
|
|
265
273
|
- **The budget is enforced, not merely reported.** A synchronous `while(true){}` never
|
|
@@ -271,16 +279,15 @@ exactly three things the previous in-process evaluator could not:
|
|
|
271
279
|
- **The script's mistakes stay the script's.** An uncaught throw, and `process.exit()`, end
|
|
272
280
|
the realm and come back as a `422` — neither reaches the runner.
|
|
273
281
|
|
|
274
|
-
It costs about **
|
|
275
|
-
laptop), a 200 KiB body about
|
|
276
|
-
types on every realm — precompiling it to JavaScript would buy that back, and
|
|
277
|
-
not done: a build artefact that goes stale against its source fails silently,
|
|
278
|
-
is the one where a wrong line number is invisible.
|
|
282
|
+
It costs about **27ms per execution** end to end for a trivial script (Node 24, M-series
|
|
283
|
+
laptop), a 200 KiB body about 30ms. Roughly 11ms of that is Node re-stripping `realm.ts`'s
|
|
284
|
+
types on every realm — precompiling it to JavaScript measures 17ms and would buy that back, and
|
|
285
|
+
is deliberately not done: a build artefact that goes stale against its source fails silently,
|
|
286
|
+
and this file is the one where a wrong line number is invisible.
|
|
279
287
|
|
|
280
|
-
|
|
281
|
-
|
|
282
|
-
|
|
283
|
-
(below), so it is the live upgrade path rather than a theoretical one.
|
|
288
|
+
A subprocess per execution measures ~55ms here, twice the thread rather than the ten times it
|
|
289
|
+
cost on the previous runtime. It contains the two things a thread cannot (below), so the upgrade
|
|
290
|
+
path is a live one at a price worth naming rather than a theoretical one.
|
|
284
291
|
|
|
285
292
|
## What this is not
|
|
286
293
|
|
|
@@ -291,7 +298,7 @@ thread no longer wins on price, and a subprocess contains the two things a threa
|
|
|
291
298
|
it needed was surface with nothing behind it. A value that must survive a retry belongs in
|
|
292
299
|
the definition, passed through `input`.
|
|
293
300
|
- **Not a sandbox.** The realm isolates *execution*, not *authority*: a script gets the
|
|
294
|
-
worker's filesystem, network and environment, and
|
|
301
|
+
worker's filesystem, network and environment, and any node builtin it imports. That is
|
|
295
302
|
deliberate — a script task is meant to do real work — but the trust boundary stays the
|
|
296
303
|
same-trust-domain one (your genroc, your worker host). It is not the multi-tenant story, and
|
|
297
304
|
nothing here should be mistaken for one. Pulling does move the boundary in one useful way:
|
|
@@ -305,9 +312,9 @@ thread no longer wins on price, and a subprocess contains the two things a threa
|
|
|
305
312
|
- **Concurrency is capped by this worker, not by genroc.** Each evaluation is a thread, and
|
|
306
313
|
`CONCURRENCY` is how many it will claim at once. Raising it past what the host can run turns
|
|
307
314
|
a queue back into threads fighting over a core.
|
|
308
|
-
- **Not where imports and type checking happen.** The evaluator
|
|
309
|
-
|
|
310
|
-
|
|
315
|
+
- **Not where imports and type checking happen.** The evaluator takes one self-contained
|
|
316
|
+
module and knows nothing about TypeScript; `import.ts` is what bundles an author's script and
|
|
317
|
+
its dependencies into that module, at author time. See above.
|
|
311
318
|
- **`eval.ts` and `realm.ts` know nothing about genroc.** `worker.ts` is the entire
|
|
312
319
|
queue-facing half, which is what keeps the containment strategy swappable — and what lets
|
|
313
320
|
the realm's own properties be tested by calling `evaluate()` directly.
|
package/dist/eval.js
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
// Evaluation core:
|
|
2
|
-
// the failure kinds in README.md. Nothing here knows about
|
|
3
|
-
// thing that talks to the queue — so this stays testable and
|
|
1
|
+
// Evaluation core: import a script module in its OWN realm, call its default export, and
|
|
2
|
+
// classify every outcome into one of the failure kinds in README.md. Nothing here knows about
|
|
3
|
+
// genroc — worker.ts is the only thing that talks to the queue — so this stays testable and
|
|
4
|
+
// the containment stays swappable.
|
|
4
5
|
//
|
|
5
6
|
// The containment is a Worker per execution (realm.ts). It is what makes the budget real:
|
|
6
7
|
// a synchronous busy loop never yields, so no in-process timer can interrupt it, and only a
|
package/dist/import.js
CHANGED
|
@@ -274,12 +274,12 @@ function tsResolve(configPath) {
|
|
|
274
274
|
},
|
|
275
275
|
};
|
|
276
276
|
}
|
|
277
|
-
/** Bundles to
|
|
278
|
-
*
|
|
279
|
-
*
|
|
277
|
+
/** Bundles to a self-contained ES module, which is what the evaluator imports: the default
|
|
278
|
+
* export it calls is the author's own, so nothing wraps or rewrites the code between the two.
|
|
279
|
+
* Bundling is entirely the importer's job, so a definition version pins its code forever. */
|
|
280
280
|
async function bundle(site, root) {
|
|
281
|
-
// Builtins are EXTERNALISED as
|
|
282
|
-
// unresolved is a REFUSAL, not an external: rollup's default is to leave it as
|
|
281
|
+
// Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
|
|
282
|
+
// unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
|
|
283
283
|
// of a module that will not be there, which bundles clean and fails at runtime.
|
|
284
284
|
const built = await rollup({
|
|
285
285
|
input: site.path,
|
|
@@ -299,18 +299,14 @@ async function bundle(site, root) {
|
|
|
299
299
|
}
|
|
300
300
|
},
|
|
301
301
|
}).catch((e) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
|
|
302
|
-
const { output } = await built.generate({ format: "
|
|
302
|
+
const { output } = await built.generate({ format: "es", inlineDynamicImports: true });
|
|
303
303
|
await built.close();
|
|
304
|
-
|
|
305
|
-
|
|
306
|
-
|
|
307
|
-
|
|
308
|
-
|
|
309
|
-
|
|
310
|
-
` throw new Error(${JSON.stringify(`${site.path} has no default export function`)});`,
|
|
311
|
-
"}",
|
|
312
|
-
"return await __genroc_main(input);",
|
|
313
|
-
].join("\n");
|
|
304
|
+
// Refused here rather than in the realm: the evaluator can only report it against a running
|
|
305
|
+
// instance, and the file it names is on this machine.
|
|
306
|
+
if (!output[0].exports.includes("default")) {
|
|
307
|
+
die(`${site.path}: a script must \`export default\` the function to run`);
|
|
308
|
+
}
|
|
309
|
+
return output[0].code;
|
|
314
310
|
}
|
|
315
311
|
// ── main ───────────────────────────────────────────────────────────────────────
|
|
316
312
|
const stdin = await new Promise((resolve, reject) => {
|
package/dist/realm.js
CHANGED
|
@@ -2,80 +2,46 @@
|
|
|
2
2
|
// thread the host can kill mid-loop — the only thing that bounds a synchronous busy loop.
|
|
3
3
|
// eval.ts owns the budget and does the killing; nothing here knows about time.
|
|
4
4
|
//
|
|
5
|
-
// Everything that touches the script's VALUE lives on this side of the boundary —
|
|
5
|
+
// Everything that touches the script's VALUE lives on this side of the boundary — loading,
|
|
6
6
|
// classifying, serialising — because this is the only realm the value exists in.
|
|
7
|
-
|
|
7
|
+
var __rewriteRelativeImportExtension = (this && this.__rewriteRelativeImportExtension) || function (path, preserveJsx) {
|
|
8
|
+
if (typeof path === "string" && /^\.\.?\//.test(path)) {
|
|
9
|
+
return path.replace(/\.(tsx)$|((?:\.d)?)((?:\.[^./]+?)?)\.([cm]?)ts$/i, function (m, tsx, d, ext, cm) {
|
|
10
|
+
return tsx ? preserveJsx ? ".jsx" : ".js" : d && (!ext || !cm) ? m : (d + ext + "." + cm.toLowerCase() + "js");
|
|
11
|
+
});
|
|
12
|
+
}
|
|
13
|
+
return path;
|
|
14
|
+
};
|
|
15
|
+
import { registerHooks } from "node:module";
|
|
8
16
|
import { parentPort } from "node:worker_threads";
|
|
9
|
-
|
|
10
|
-
|
|
11
|
-
//
|
|
12
|
-
|
|
13
|
-
// `require` in scope. Passing one in is what makes an import of a builtin work at runtime rather than at
|
|
14
|
-
// typecheck only. Resolution is anchored here, which is right: only builtins reach it.
|
|
15
|
-
const scriptRequire = createRequire(import.meta.url);
|
|
17
|
+
// The script is IMPORTED as a module under a URL of our own, not compiled from a string: its
|
|
18
|
+
// frames then carry the author's own line numbers, and an `import` of a node builtin resolves
|
|
19
|
+
// the way it does everywhere else.
|
|
20
|
+
const SCRIPT_URL = "script:main";
|
|
16
21
|
const STACK_BYTES = 2_048;
|
|
17
|
-
|
|
18
|
-
|
|
19
|
-
|
|
20
|
-
|
|
21
|
-
|
|
22
|
-
|
|
23
|
-
|
|
24
|
-
|
|
25
|
-
|
|
26
|
-
|
|
27
|
-
return 0;
|
|
28
|
-
}
|
|
29
|
-
catch (err) {
|
|
30
|
-
return reportedLine(err) - 1;
|
|
31
|
-
}
|
|
32
|
-
})();
|
|
33
|
-
// V8 marks a frame compiled by the AsyncFunction constructor with the site that CALLED the
|
|
34
|
-
// constructor, then the script's OWN position:
|
|
35
|
-
// at inner (eval at run (file:///…/worker.ts:107:10), <anonymous>:6:9)
|
|
36
|
-
// `eval at` is therefore what separates script frames from runner plumbing — matched without
|
|
37
|
-
// the function name, which is whatever encloses the `new AsyncFunction` below. The LAST such
|
|
38
|
-
// frame is the body's top level: frames interleave, since a script can throw inside a native
|
|
39
|
-
// callback.
|
|
40
|
-
const SCRIPT_FRAME = /\(eval at /;
|
|
41
|
-
// The trailing `<anonymous>:LINE:COL` — the script's position, after the host file's own.
|
|
42
|
-
const POSITION = /<anonymous>:(\d+):(\d+)\)?\s*$/;
|
|
43
|
-
// ` at name (` — absent on the top-level frame, which V8 names `eval`.
|
|
44
|
-
const FRAME_NAME = /^\s*at\s+(?:async\s+)?([^\s(]+)\s*\(/;
|
|
45
|
-
/** Line number of the throw as the engine reported it, or 1 if the stack is unreadable. */
|
|
46
|
-
function reportedLine(err) {
|
|
47
|
-
const stack = err instanceof Error && typeof err.stack === "string" ? err.stack : "";
|
|
48
|
-
const frame = stack.split("\n").find((l) => SCRIPT_FRAME.test(l)) ?? "";
|
|
49
|
-
const m = frame.match(POSITION);
|
|
50
|
-
return m ? Number(m[1]) : 1;
|
|
51
|
-
}
|
|
52
|
-
/** Renumbers each script frame to the line the AUTHOR wrote and drops the runner's own.
|
|
53
|
-
* Rewriting the whole location is also what keeps the runner's path out of a script's
|
|
54
|
-
* stack — V8 puts it inside every compiled frame. */
|
|
55
|
-
function scriptStack(err, offset) {
|
|
22
|
+
// The only channel a load hook has to the source. Written per request and read once — a realm
|
|
23
|
+
// evaluates one script and is then discarded, so no second execution can observe it.
|
|
24
|
+
let source = "";
|
|
25
|
+
registerHooks({
|
|
26
|
+
resolve: (specifier, context, next) => specifier === SCRIPT_URL ? { url: SCRIPT_URL, shortCircuit: true } : next(specifier, context),
|
|
27
|
+
load: (url, context, next) => url === SCRIPT_URL ? { format: "module", source, shortCircuit: true } : next(url, context),
|
|
28
|
+
});
|
|
29
|
+
/** Keeps the frames that are the script's own. Everything below the last of them is runner
|
|
30
|
+
* plumbing the author cannot act on, and V8 puts this file's path in it. */
|
|
31
|
+
function scriptStack(err) {
|
|
56
32
|
if (!(err instanceof Error) || typeof err.stack !== "string")
|
|
57
33
|
return undefined;
|
|
58
34
|
const lines = err.stack.split("\n");
|
|
59
|
-
let
|
|
35
|
+
let last = 0;
|
|
60
36
|
for (let i = 0; i < lines.length; i++)
|
|
61
|
-
if (
|
|
62
|
-
|
|
63
|
-
const
|
|
64
|
-
|
|
65
|
-
const pos = line.match(POSITION);
|
|
66
|
-
if (!pos)
|
|
67
|
-
return line; // the `Error: message` header and native frames, kept as-is
|
|
68
|
-
const name = line.match(FRAME_NAME)?.[1];
|
|
69
|
-
const at = name && name !== "eval" && name !== "anonymous" ? `at ${name} ` : "at ";
|
|
70
|
-
const indent = line.match(/^\s*/)[0];
|
|
71
|
-
return `${indent}${at}(script:${Math.max(1, Number(pos[1]) - offset)}:${pos[2]})`;
|
|
72
|
-
})
|
|
73
|
-
.join("\n");
|
|
74
|
-
return frames.length > STACK_BYTES ? frames.slice(0, STACK_BYTES) : frames;
|
|
37
|
+
if (lines[i].includes(SCRIPT_URL))
|
|
38
|
+
last = i;
|
|
39
|
+
const stack = lines.slice(0, last + 1).join("\n");
|
|
40
|
+
return stack.length > STACK_BYTES ? stack.slice(0, STACK_BYTES) : stack;
|
|
75
41
|
}
|
|
76
|
-
function describe(err, kind
|
|
42
|
+
function describe(err, kind) {
|
|
77
43
|
if (err instanceof Error) {
|
|
78
|
-
return { kind, name: err.name, message: err.message, stack: scriptStack(err
|
|
44
|
+
return { kind, name: err.name, message: err.message, stack: scriptStack(err) };
|
|
79
45
|
}
|
|
80
46
|
// A script may throw a non-Error (`throw {code: "x"}`), so name/message must not assume one.
|
|
81
47
|
return { kind, name: "Thrown", message: safeText(err) };
|
|
@@ -88,23 +54,39 @@ function safeText(v) {
|
|
|
88
54
|
return String(v);
|
|
89
55
|
}
|
|
90
56
|
}
|
|
57
|
+
/** A module that will not parse, or names an import nothing resolves, is broken code — only
|
|
58
|
+
* editing it helps. Anything else thrown by the import is the module's top level running,
|
|
59
|
+
* which is the script throwing. */
|
|
60
|
+
function unloadable(err) {
|
|
61
|
+
const code = err?.code;
|
|
62
|
+
return err instanceof SyntaxError || (typeof code === "string" && code.startsWith("ERR_MODULE"));
|
|
63
|
+
}
|
|
91
64
|
async function run(req) {
|
|
92
|
-
|
|
93
|
-
let
|
|
65
|
+
source = req.code;
|
|
66
|
+
let main;
|
|
94
67
|
try {
|
|
95
|
-
|
|
96
|
-
// never be hit. Repeated compilation is the price of the fresh global object.
|
|
97
|
-
fn = new AsyncFunction("input", "require", STRICT + req.code);
|
|
68
|
+
main = (await import(__rewriteRelativeImportExtension(SCRIPT_URL))).default;
|
|
98
69
|
}
|
|
99
70
|
catch (err) {
|
|
100
|
-
return { ok: false, failure: describe(err, "compile_error"
|
|
71
|
+
return { ok: false, failure: describe(err, unloadable(err) ? "compile_error" : "threw") };
|
|
72
|
+
}
|
|
73
|
+
if (typeof main !== "function") {
|
|
74
|
+
const got = main === undefined ? "no default export" : `a ${typeof main}`;
|
|
75
|
+
return {
|
|
76
|
+
ok: false,
|
|
77
|
+
failure: {
|
|
78
|
+
kind: "compile_error",
|
|
79
|
+
name: "NoDefaultExport",
|
|
80
|
+
message: `a script must export default a function; this one has ${got}`,
|
|
81
|
+
},
|
|
82
|
+
};
|
|
101
83
|
}
|
|
102
84
|
let value;
|
|
103
85
|
try {
|
|
104
|
-
value = await
|
|
86
|
+
value = await main(req.input);
|
|
105
87
|
}
|
|
106
88
|
catch (err) {
|
|
107
|
-
return { ok: false, failure: describe(err, "threw"
|
|
89
|
+
return { ok: false, failure: describe(err, "threw") };
|
|
108
90
|
}
|
|
109
91
|
try {
|
|
110
92
|
// undefined stringifies to undefined, not "undefined"; an empty body is how genroc
|
|
@@ -112,11 +94,20 @@ async function run(req) {
|
|
|
112
94
|
return { ok: true, body: value === undefined ? "" : JSON.stringify(value) ?? "" };
|
|
113
95
|
}
|
|
114
96
|
catch (err) {
|
|
115
|
-
return { ok: false, failure: describe(err, "nonserializable"
|
|
97
|
+
return { ok: false, failure: describe(err, "nonserializable") };
|
|
116
98
|
}
|
|
117
99
|
}
|
|
100
|
+
// The realm's stdio is a pipe to the host thread, and eval.ts terminates this thread the moment
|
|
101
|
+
// the reply lands — so whatever a script wrote last is still in the pipe when it dies. An empty
|
|
102
|
+
// write's callback fires once the queue ahead of it has drained, which makes the reply a barrier
|
|
103
|
+
// for `console` and a direct process.stdout.write alike: both are this stream.
|
|
104
|
+
function flush(stream) {
|
|
105
|
+
return new Promise((resolve) => stream.write("", () => resolve()));
|
|
106
|
+
}
|
|
118
107
|
// Non-null because this module only ever runs as a worker entry point; a null port here
|
|
119
108
|
// would mean eval.ts loaded it as a plain module, which nothing does.
|
|
120
109
|
parentPort.on("message", async (req) => {
|
|
121
|
-
|
|
110
|
+
const reply = await run(req);
|
|
111
|
+
await Promise.all([flush(process.stdout), flush(process.stderr)]);
|
|
112
|
+
parentPort.postMessage(reply);
|
|
122
113
|
});
|
package/dist/worker.js
CHANGED
|
@@ -4,6 +4,7 @@
|
|
|
4
4
|
// nothing about the queue, which is what keeps the containment strategy swappable.
|
|
5
5
|
//
|
|
6
6
|
// See README.md for the contract, and specs/external-task-queue.md for the queue itself.
|
|
7
|
+
import { readFileSync } from "node:fs";
|
|
7
8
|
import { evaluate } from "./eval.js";
|
|
8
9
|
const SERVER = (process.env.GENROC_SERVER ?? "http://localhost:8448").replace(/\/$/, "");
|
|
9
10
|
const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
|
|
@@ -15,7 +16,11 @@ const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
|
|
|
15
16
|
// Sent as a header rather than in the URL because Node's fetch REFUSES a URL carrying
|
|
16
17
|
// credentials ("Request cannot be constructed from a URL that includes credentials"), so the
|
|
17
18
|
// basic-auth-in-the-URL trick that works for genctl is not available here.
|
|
18
|
-
|
|
19
|
+
// GENROC_TOKEN_FILE is the mounted-secret shape: a credential in a file rather than an
|
|
20
|
+
// environment variable, so it stays out of `docker inspect` and out of the process environment
|
|
21
|
+
// any child inherits. The inline variable wins when both are set.
|
|
22
|
+
const TOKEN = process.env.GENROC_TOKEN ??
|
|
23
|
+
(process.env.GENROC_TOKEN_FILE ? readFileSync(process.env.GENROC_TOKEN_FILE, "utf8").trim() : "");
|
|
19
24
|
const authHeaders = TOKEN ? { authorization: `Bearer ${TOKEN}` } : {};
|
|
20
25
|
// Concurrency is the worker's to set, and that is the point of pulling: under the old fetch
|
|
21
26
|
// shape genroc decided how many scripts ran at once (--max-concurrent, default 200) and the
|
package/eval.ts
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
// Evaluation core:
|
|
2
|
-
// the failure kinds in README.md. Nothing here knows about
|
|
3
|
-
// thing that talks to the queue — so this stays testable and
|
|
1
|
+
// Evaluation core: import a script module in its OWN realm, call its default export, and
|
|
2
|
+
// classify every outcome into one of the failure kinds in README.md. Nothing here knows about
|
|
3
|
+
// genroc — worker.ts is the only thing that talks to the queue — so this stays testable and
|
|
4
|
+
// the containment stays swappable.
|
|
4
5
|
//
|
|
5
6
|
// The containment is a Worker per execution (realm.ts). It is what makes the budget real:
|
|
6
7
|
// a synchronous busy loop never yields, so no in-process timer can interrupt it, and only a
|
|
@@ -9,6 +10,7 @@
|
|
|
9
10
|
import { Worker } from "node:worker_threads";
|
|
10
11
|
|
|
11
12
|
export type EvalRequest = {
|
|
13
|
+
/** An ES module whose default export is the function to run; `input` is its argument. */
|
|
12
14
|
code: string;
|
|
13
15
|
input?: unknown;
|
|
14
16
|
timeout_ms?: number;
|
package/import.ts
CHANGED
|
@@ -309,12 +309,12 @@ function tsResolve(configPath: string | null): Plugin {
|
|
|
309
309
|
};
|
|
310
310
|
}
|
|
311
311
|
|
|
312
|
-
/** Bundles to
|
|
313
|
-
*
|
|
314
|
-
*
|
|
312
|
+
/** Bundles to a self-contained ES module, which is what the evaluator imports: the default
|
|
313
|
+
* export it calls is the author's own, so nothing wraps or rewrites the code between the two.
|
|
314
|
+
* Bundling is entirely the importer's job, so a definition version pins its code forever. */
|
|
315
315
|
async function bundle(site: Site, root: string): Promise<string> {
|
|
316
|
-
// Builtins are EXTERNALISED as
|
|
317
|
-
// unresolved is a REFUSAL, not an external: rollup's default is to leave it as
|
|
316
|
+
// Builtins are EXTERNALISED as imports the realm resolves natively. Anything else
|
|
317
|
+
// unresolved is a REFUSAL, not an external: rollup's default is to leave it as an import
|
|
318
318
|
// of a module that will not be there, which bundles clean and fails at runtime.
|
|
319
319
|
const built = await rollup({
|
|
320
320
|
input: site.path,
|
|
@@ -335,18 +335,14 @@ async function bundle(site: Site, root: string): Promise<string> {
|
|
|
335
335
|
},
|
|
336
336
|
}).catch((e: unknown) => die(`${site.path}: ${e instanceof Error ? e.message : String(e)}`));
|
|
337
337
|
|
|
338
|
-
const { output } = await built.generate({ format: "
|
|
338
|
+
const { output } = await built.generate({ format: "es", inlineDynamicImports: true });
|
|
339
339
|
await built.close();
|
|
340
|
-
|
|
341
|
-
|
|
342
|
-
|
|
343
|
-
|
|
344
|
-
|
|
345
|
-
|
|
346
|
-
` throw new Error(${JSON.stringify(`${site.path} has no default export function`)});`,
|
|
347
|
-
"}",
|
|
348
|
-
"return await __genroc_main(input);",
|
|
349
|
-
].join("\n");
|
|
340
|
+
// Refused here rather than in the realm: the evaluator can only report it against a running
|
|
341
|
+
// instance, and the file it names is on this machine.
|
|
342
|
+
if (!output[0].exports.includes("default")) {
|
|
343
|
+
die(`${site.path}: a script must \`export default\` the function to run`);
|
|
344
|
+
}
|
|
345
|
+
return output[0].code;
|
|
350
346
|
}
|
|
351
347
|
|
|
352
348
|
// ── main ───────────────────────────────────────────────────────────────────────
|
package/package.json
CHANGED
package/realm.ts
CHANGED
|
@@ -2,88 +2,45 @@
|
|
|
2
2
|
// thread the host can kill mid-loop — the only thing that bounds a synchronous busy loop.
|
|
3
3
|
// eval.ts owns the budget and does the killing; nothing here knows about time.
|
|
4
4
|
//
|
|
5
|
-
// Everything that touches the script's VALUE lives on this side of the boundary —
|
|
5
|
+
// Everything that touches the script's VALUE lives on this side of the boundary — loading,
|
|
6
6
|
// classifying, serialising — because this is the only realm the value exists in.
|
|
7
7
|
|
|
8
|
-
import {
|
|
8
|
+
import { registerHooks } from "node:module";
|
|
9
9
|
import { parentPort } from "node:worker_threads";
|
|
10
10
|
|
|
11
11
|
import type { EvalFailure, FailureKind, WorkerReply, WorkerRequest } from "./eval.ts";
|
|
12
12
|
|
|
13
|
-
|
|
14
|
-
|
|
15
|
-
|
|
16
|
-
|
|
17
|
-
const STRICT = '"use strict";\n';
|
|
18
|
-
|
|
19
|
-
// Bundled `node:*` imports survive as `require` calls — the importer externalises builtins and
|
|
20
|
-
// inlines everything else — and a function built by the AsyncFunction constructor has no
|
|
21
|
-
// `require` in scope. Passing one in is what makes an import of a builtin work at runtime rather than at
|
|
22
|
-
// typecheck only. Resolution is anchored here, which is right: only builtins reach it.
|
|
23
|
-
const scriptRequire = createRequire(import.meta.url);
|
|
13
|
+
// The script is IMPORTED as a module under a URL of our own, not compiled from a string: its
|
|
14
|
+
// frames then carry the author's own line numbers, and an `import` of a node builtin resolves
|
|
15
|
+
// the way it does everywhere else.
|
|
16
|
+
const SCRIPT_URL = "script:main";
|
|
24
17
|
const STACK_BYTES = 2_048;
|
|
25
18
|
|
|
26
|
-
|
|
27
|
-
|
|
28
|
-
|
|
29
|
-
* script's error location the day it changes.
|
|
30
|
-
*/
|
|
31
|
-
const lineOffset: Promise<number> = (async () => {
|
|
32
|
-
// Same parameter list as a real compile: the preamble is what is being measured.
|
|
33
|
-
const probe = new AsyncFunction("input", "require", STRICT + "throw new Error('probe');");
|
|
34
|
-
try {
|
|
35
|
-
await probe();
|
|
36
|
-
return 0;
|
|
37
|
-
} catch (err) {
|
|
38
|
-
return reportedLine(err) - 1;
|
|
39
|
-
}
|
|
40
|
-
})();
|
|
41
|
-
|
|
42
|
-
// V8 marks a frame compiled by the AsyncFunction constructor with the site that CALLED the
|
|
43
|
-
// constructor, then the script's OWN position:
|
|
44
|
-
// at inner (eval at run (file:///…/worker.ts:107:10), <anonymous>:6:9)
|
|
45
|
-
// `eval at` is therefore what separates script frames from runner plumbing — matched without
|
|
46
|
-
// the function name, which is whatever encloses the `new AsyncFunction` below. The LAST such
|
|
47
|
-
// frame is the body's top level: frames interleave, since a script can throw inside a native
|
|
48
|
-
// callback.
|
|
49
|
-
const SCRIPT_FRAME = /\(eval at /;
|
|
50
|
-
// The trailing `<anonymous>:LINE:COL` — the script's position, after the host file's own.
|
|
51
|
-
const POSITION = /<anonymous>:(\d+):(\d+)\)?\s*$/;
|
|
52
|
-
// ` at name (` — absent on the top-level frame, which V8 names `eval`.
|
|
53
|
-
const FRAME_NAME = /^\s*at\s+(?:async\s+)?([^\s(]+)\s*\(/;
|
|
19
|
+
// The only channel a load hook has to the source. Written per request and read once — a realm
|
|
20
|
+
// evaluates one script and is then discarded, so no second execution can observe it.
|
|
21
|
+
let source = "";
|
|
54
22
|
|
|
55
|
-
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
|
|
61
|
-
}
|
|
23
|
+
registerHooks({
|
|
24
|
+
resolve: (specifier, context, next) =>
|
|
25
|
+
specifier === SCRIPT_URL ? { url: SCRIPT_URL, shortCircuit: true } : next(specifier, context),
|
|
26
|
+
load: (url, context, next) =>
|
|
27
|
+
url === SCRIPT_URL ? { format: "module", source, shortCircuit: true } : next(url, context),
|
|
28
|
+
});
|
|
62
29
|
|
|
63
|
-
/**
|
|
64
|
-
*
|
|
65
|
-
|
|
66
|
-
function scriptStack(err: unknown, offset: number): string | undefined {
|
|
30
|
+
/** Keeps the frames that are the script's own. Everything below the last of them is runner
|
|
31
|
+
* plumbing the author cannot act on, and V8 puts this file's path in it. */
|
|
32
|
+
function scriptStack(err: unknown): string | undefined {
|
|
67
33
|
if (!(err instanceof Error) || typeof err.stack !== "string") return undefined;
|
|
68
34
|
const lines = err.stack.split("\n");
|
|
69
|
-
let
|
|
70
|
-
for (let i = 0; i < lines.length; i++) if (
|
|
71
|
-
const
|
|
72
|
-
|
|
73
|
-
const pos = line.match(POSITION);
|
|
74
|
-
if (!pos) return line; // the `Error: message` header and native frames, kept as-is
|
|
75
|
-
const name = line.match(FRAME_NAME)?.[1];
|
|
76
|
-
const at = name && name !== "eval" && name !== "anonymous" ? `at ${name} ` : "at ";
|
|
77
|
-
const indent = line.match(/^\s*/)![0];
|
|
78
|
-
return `${indent}${at}(script:${Math.max(1, Number(pos[1]) - offset)}:${pos[2]})`;
|
|
79
|
-
})
|
|
80
|
-
.join("\n");
|
|
81
|
-
return frames.length > STACK_BYTES ? frames.slice(0, STACK_BYTES) : frames;
|
|
35
|
+
let last = 0;
|
|
36
|
+
for (let i = 0; i < lines.length; i++) if (lines[i]!.includes(SCRIPT_URL)) last = i;
|
|
37
|
+
const stack = lines.slice(0, last + 1).join("\n");
|
|
38
|
+
return stack.length > STACK_BYTES ? stack.slice(0, STACK_BYTES) : stack;
|
|
82
39
|
}
|
|
83
40
|
|
|
84
|
-
function describe(err: unknown, kind: FailureKind
|
|
41
|
+
function describe(err: unknown, kind: FailureKind): EvalFailure {
|
|
85
42
|
if (err instanceof Error) {
|
|
86
|
-
return { kind, name: err.name, message: err.message, stack: scriptStack(err
|
|
43
|
+
return { kind, name: err.name, message: err.message, stack: scriptStack(err) };
|
|
87
44
|
}
|
|
88
45
|
// A script may throw a non-Error (`throw {code: "x"}`), so name/message must not assume one.
|
|
89
46
|
return { kind, name: "Thrown", message: safeText(err) };
|
|
@@ -97,23 +54,40 @@ function safeText(v: unknown): string {
|
|
|
97
54
|
}
|
|
98
55
|
}
|
|
99
56
|
|
|
57
|
+
/** A module that will not parse, or names an import nothing resolves, is broken code — only
|
|
58
|
+
* editing it helps. Anything else thrown by the import is the module's top level running,
|
|
59
|
+
* which is the script throwing. */
|
|
60
|
+
function unloadable(err: unknown): boolean {
|
|
61
|
+
const code = (err as { code?: unknown } | null)?.code;
|
|
62
|
+
return err instanceof SyntaxError || (typeof code === "string" && code.startsWith("ERR_MODULE"));
|
|
63
|
+
}
|
|
64
|
+
|
|
100
65
|
async function run(req: WorkerRequest): Promise<WorkerReply> {
|
|
101
|
-
|
|
66
|
+
source = req.code;
|
|
102
67
|
|
|
103
|
-
let
|
|
68
|
+
let main: unknown;
|
|
104
69
|
try {
|
|
105
|
-
|
|
106
|
-
// never be hit. Repeated compilation is the price of the fresh global object.
|
|
107
|
-
fn = new AsyncFunction("input", "require", STRICT + req.code);
|
|
70
|
+
main = (await import(SCRIPT_URL)).default;
|
|
108
71
|
} catch (err) {
|
|
109
|
-
return { ok: false, failure: describe(err, "compile_error"
|
|
72
|
+
return { ok: false, failure: describe(err, unloadable(err) ? "compile_error" : "threw") };
|
|
73
|
+
}
|
|
74
|
+
if (typeof main !== "function") {
|
|
75
|
+
const got = main === undefined ? "no default export" : `a ${typeof main}`;
|
|
76
|
+
return {
|
|
77
|
+
ok: false,
|
|
78
|
+
failure: {
|
|
79
|
+
kind: "compile_error",
|
|
80
|
+
name: "NoDefaultExport",
|
|
81
|
+
message: `a script must export default a function; this one has ${got}`,
|
|
82
|
+
},
|
|
83
|
+
};
|
|
110
84
|
}
|
|
111
85
|
|
|
112
86
|
let value: unknown;
|
|
113
87
|
try {
|
|
114
|
-
value = await
|
|
88
|
+
value = await (main as (input: unknown) => unknown)(req.input);
|
|
115
89
|
} catch (err) {
|
|
116
|
-
return { ok: false, failure: describe(err, "threw"
|
|
90
|
+
return { ok: false, failure: describe(err, "threw") };
|
|
117
91
|
}
|
|
118
92
|
|
|
119
93
|
try {
|
|
@@ -121,12 +95,22 @@ async function run(req: WorkerRequest): Promise<WorkerReply> {
|
|
|
121
95
|
// spells null, which is the right reading of a script that returned nothing.
|
|
122
96
|
return { ok: true, body: value === undefined ? "" : JSON.stringify(value) ?? "" };
|
|
123
97
|
} catch (err) {
|
|
124
|
-
return { ok: false, failure: describe(err, "nonserializable"
|
|
98
|
+
return { ok: false, failure: describe(err, "nonserializable") };
|
|
125
99
|
}
|
|
126
100
|
}
|
|
127
101
|
|
|
102
|
+
// The realm's stdio is a pipe to the host thread, and eval.ts terminates this thread the moment
|
|
103
|
+
// the reply lands — so whatever a script wrote last is still in the pipe when it dies. An empty
|
|
104
|
+
// write's callback fires once the queue ahead of it has drained, which makes the reply a barrier
|
|
105
|
+
// for `console` and a direct process.stdout.write alike: both are this stream.
|
|
106
|
+
function flush(stream: NodeJS.WriteStream): Promise<void> {
|
|
107
|
+
return new Promise((resolve) => stream.write("", () => resolve()));
|
|
108
|
+
}
|
|
109
|
+
|
|
128
110
|
// Non-null because this module only ever runs as a worker entry point; a null port here
|
|
129
111
|
// would mean eval.ts loaded it as a plain module, which nothing does.
|
|
130
112
|
parentPort!.on("message", async (req: WorkerRequest) => {
|
|
131
|
-
|
|
113
|
+
const reply = await run(req);
|
|
114
|
+
await Promise.all([flush(process.stdout), flush(process.stderr)]);
|
|
115
|
+
parentPort!.postMessage(reply);
|
|
132
116
|
});
|
package/worker.ts
CHANGED
|
@@ -5,6 +5,7 @@
|
|
|
5
5
|
//
|
|
6
6
|
// See README.md for the contract, and specs/external-task-queue.md for the queue itself.
|
|
7
7
|
|
|
8
|
+
import { readFileSync } from "node:fs";
|
|
8
9
|
import { evaluate, type EvalRequest, type FailureKind } from "./eval.ts";
|
|
9
10
|
|
|
10
11
|
const SERVER = (process.env.GENROC_SERVER ?? "http://localhost:8448").replace(/\/$/, "");
|
|
@@ -17,7 +18,12 @@ const WORKER_ID = process.env.WORKER_ID ?? `evaluator-${process.pid}`;
|
|
|
17
18
|
// Sent as a header rather than in the URL because Node's fetch REFUSES a URL carrying
|
|
18
19
|
// credentials ("Request cannot be constructed from a URL that includes credentials"), so the
|
|
19
20
|
// basic-auth-in-the-URL trick that works for genctl is not available here.
|
|
20
|
-
|
|
21
|
+
// GENROC_TOKEN_FILE is the mounted-secret shape: a credential in a file rather than an
|
|
22
|
+
// environment variable, so it stays out of `docker inspect` and out of the process environment
|
|
23
|
+
// any child inherits. The inline variable wins when both are set.
|
|
24
|
+
const TOKEN =
|
|
25
|
+
process.env.GENROC_TOKEN ??
|
|
26
|
+
(process.env.GENROC_TOKEN_FILE ? readFileSync(process.env.GENROC_TOKEN_FILE, "utf8").trim() : "");
|
|
21
27
|
const authHeaders: Record<string, string> = TOKEN ? { authorization: `Bearer ${TOKEN}` } : {};
|
|
22
28
|
// Concurrency is the worker's to set, and that is the point of pulling: under the old fetch
|
|
23
29
|
// shape genroc decided how many scripts ran at once (--max-concurrent, default 200) and the
|