@velum-labs/routekit-gateway 0.18.1 → 0.18.2

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/dist/backend.d.ts CHANGED
@@ -44,8 +44,8 @@ export type Backend = {
44
44
  resolveModelRoute?(requested: string | undefined, nativeProvider?: string): BackendModelRoute | undefined;
45
45
  /**
46
46
  * Whether the backend serves this exact model id itself. Unlike
47
- * {@link resolveModel} — which folds unknown
48
- * ids into the default — this distinguishes "mine" from "unknown", so the
47
+ * The defaulting resolver folds unknown ids into the default; this
48
+ * distinguishes "mine" from "unknown", so the
49
49
  * gateway can hand unknown ids to a relay (e.g. the Codex backend relay)
50
50
  * instead of silently routing them to the default.
51
51
  */
@@ -5,6 +5,7 @@ export type SwitchingGatewayProxy = {
5
5
  target(): string;
6
6
  swapTarget(target: string): string;
7
7
  waitForTargetIdle(target: string, graceMs: number): Promise<boolean>;
8
+ retire(graceMs?: number): Promise<void>;
8
9
  drain(graceMs?: number): Promise<void>;
9
10
  close(): Promise<void>;
10
11
  };
@@ -68,13 +68,17 @@ function writeJson(res, status, value) {
68
68
  res.setHeader("content-length", String(payload.length));
69
69
  res.end(payload);
70
70
  }
71
- async function pipe(res, upstream) {
71
+ async function pipe(res, upstream, closeConnection) {
72
72
  res.statusCode = upstream.status;
73
73
  for (const [name, value] of upstream.headers) {
74
74
  if (!HOP_BY_HOP.has(name.toLowerCase()) && name.toLowerCase() !== "content-length") {
75
75
  res.setHeader(name, value);
76
76
  }
77
77
  }
78
+ if (closeConnection()) {
79
+ res.shouldKeepAlive = false;
80
+ res.setHeader("connection", "close");
81
+ }
78
82
  if (upstream.body === null) {
79
83
  res.end();
80
84
  return;
@@ -108,8 +112,13 @@ export async function startSwitchingGatewayProxy(input) {
108
112
  };
109
113
  const generations = new Map([[active.url, active]]);
110
114
  let draining = false;
115
+ let retiring = false;
111
116
  let inflight = 0;
112
117
  const server = createServer((req, res) => {
118
+ if (retiring) {
119
+ res.shouldKeepAlive = false;
120
+ res.setHeader("connection", "close");
121
+ }
113
122
  inflight += 1;
114
123
  res.once("close", () => {
115
124
  inflight -= 1;
@@ -170,7 +179,7 @@ export async function startSwitchingGatewayProxy(input) {
170
179
  AbortSignal.timeout(10 * 60 * 1000)
171
180
  ])
172
181
  });
173
- await pipe(res, upstream);
182
+ await pipe(res, upstream, () => retiring);
174
183
  }
175
184
  catch {
176
185
  if (!res.destroyed && !res.headersSent) {
@@ -205,14 +214,33 @@ export async function startSwitchingGatewayProxy(input) {
205
214
  const address = server.address();
206
215
  const port = typeof address === "object" && address !== null ? address.port : input.port ?? 0;
207
216
  let drainRun;
217
+ let retireRun;
218
+ const waitForInflight = async (graceMs) => {
219
+ const deadline = Date.now() + graceMs;
220
+ while (inflight > 0 && Date.now() < deadline) {
221
+ await new Promise((resolve) => setTimeout(resolve, 50));
222
+ }
223
+ };
224
+ const closeRetiringListener = async (graceMs) => {
225
+ server.closeIdleConnections();
226
+ const closed = new Promise((resolve) => server.close(() => resolve()));
227
+ await waitForInflight(graceMs);
228
+ if (inflight > 0)
229
+ server.closeAllConnections();
230
+ await closed;
231
+ };
232
+ const retire = (graceMs = 0) => {
233
+ retireRun ??= (async () => {
234
+ retiring = true;
235
+ await closeRetiringListener(graceMs);
236
+ })();
237
+ return retireRun;
238
+ };
208
239
  const drain = (graceMs = 0) => {
209
240
  drainRun ??= (async () => {
210
241
  draining = true;
211
242
  server.closeIdleConnections();
212
- const deadline = Date.now() + graceMs;
213
- while (inflight > 0 && Date.now() < deadline) {
214
- await new Promise((resolve) => setTimeout(resolve, 50));
215
- }
243
+ await waitForInflight(graceMs);
216
244
  const closed = new Promise((resolve) => server.close(() => resolve()));
217
245
  server.closeAllConnections();
218
246
  await closed;
@@ -252,6 +280,7 @@ export async function startSwitchingGatewayProxy(input) {
252
280
  generations.delete(generation.url);
253
281
  return result;
254
282
  },
283
+ retire,
255
284
  drain,
256
285
  close: async () => await drain(0)
257
286
  };
@@ -0,0 +1 @@
1
+ export {};
@@ -0,0 +1,88 @@
1
+ import assert from "node:assert/strict";
2
+ import { createServer } from "node:http";
3
+ import test from "node:test";
4
+ import { startSwitchingGatewayProxy } from "../switching-proxy.js";
5
+ async function heldTarget() {
6
+ let markStarted;
7
+ const started = new Promise((resolve) => {
8
+ markStarted = resolve;
9
+ });
10
+ let release;
11
+ const released = new Promise((resolve) => {
12
+ release = resolve;
13
+ });
14
+ const server = createServer((request, response) => {
15
+ if (request.url === "/stream") {
16
+ response.setHeader("content-type", "text/plain");
17
+ response.write("first\n");
18
+ markStarted();
19
+ void released.then(() => response.end("last\n"));
20
+ return;
21
+ }
22
+ response.end("ok");
23
+ });
24
+ await new Promise((resolve, reject) => {
25
+ server.once("error", reject);
26
+ server.listen(0, "127.0.0.1", () => resolve());
27
+ });
28
+ const port = server.address().port;
29
+ return {
30
+ url: `http://127.0.0.1:${port}`,
31
+ started,
32
+ release,
33
+ close: async () => await new Promise((resolve, reject) => server.close((error) => (error === undefined ? resolve() : reject(error))))
34
+ };
35
+ }
36
+ test("switching proxy retirement preserves an admitted streaming response", async () => {
37
+ const target = await heldTarget();
38
+ const proxy = await startSwitchingGatewayProxy({ target: target.url });
39
+ try {
40
+ const response = await fetch(`${proxy.url()}/stream`);
41
+ const reader = response.body?.getReader();
42
+ assert.ok(reader !== undefined);
43
+ const first = await reader.read();
44
+ assert.match(Buffer.from(first.value ?? []).toString("utf8"), /first/);
45
+ await target.started;
46
+ const retiring = proxy.retire(2_000);
47
+ target.release();
48
+ let remainder = "";
49
+ for (;;) {
50
+ const { done, value } = await reader.read();
51
+ if (done)
52
+ break;
53
+ remainder += Buffer.from(value ?? []).toString("utf8");
54
+ }
55
+ assert.match(remainder, /last/);
56
+ await retiring;
57
+ }
58
+ finally {
59
+ await proxy.close();
60
+ await target.close();
61
+ }
62
+ });
63
+ test("switching proxy retirement deadline terminates a stream that never finishes", async () => {
64
+ const target = await heldTarget();
65
+ const proxy = await startSwitchingGatewayProxy({ target: target.url });
66
+ try {
67
+ const response = await fetch(`${proxy.url()}/stream`);
68
+ const reader = response.body?.getReader();
69
+ assert.ok(reader !== undefined);
70
+ await reader.read();
71
+ await target.started;
72
+ const startedAt = Date.now();
73
+ await proxy.retire(250);
74
+ assert.ok(Date.now() - startedAt >= 200);
75
+ await assert.rejects(async () => {
76
+ for (;;) {
77
+ const { done } = await reader.read();
78
+ if (done)
79
+ break;
80
+ }
81
+ });
82
+ }
83
+ finally {
84
+ target.release();
85
+ await proxy.close();
86
+ await target.close();
87
+ }
88
+ });
package/package.json CHANGED
@@ -1,7 +1,7 @@
1
1
  {
2
2
  "name": "@velum-labs/routekit-gateway",
3
3
  "private": false,
4
- "version": "0.18.1",
4
+ "version": "0.18.2",
5
5
  "repository": {
6
6
  "type": "git",
7
7
  "url": "git+https://github.com/velum-labs/routekit.git",
@@ -29,10 +29,10 @@
29
29
  "@aws-sdk/client-bedrock": "3.1095.0",
30
30
  "@aws-sdk/client-bedrock-runtime": "3.1095.0",
31
31
  "zod": "4.4.3",
32
- "@velum-labs/routekit-contracts": "0.18.1",
33
- "@velum-labs/routekit-registry": "0.18.1",
34
- "@velum-labs/routekit-runtime": "0.18.1",
35
- "@velum-labs/routekit-tracing": "0.18.1"
32
+ "@velum-labs/routekit-contracts": "0.18.2",
33
+ "@velum-labs/routekit-registry": "0.18.2",
34
+ "@velum-labs/routekit-runtime": "0.18.2",
35
+ "@velum-labs/routekit-tracing": "0.18.2"
36
36
  },
37
37
  "keywords": [
38
38
  "routekit",