Skip to content

Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure

TypeScript

test(mcp-host): pin the proxy 408 request_timeout as non-retryable (#739)

Alfredo Lopez75bd6cfd866b
diff --git a/mcp-host/package-lock.json b/mcp-host/package-lock.json
index f4c6f849e..1a72211cb 100644
--- a/mcp-host/package-lock.json
+++ b/mcp-host/package-lock.json
@@ -1,12 +1,12 @@
{
"name": "clerum-mcp-host",
- "version": "0.1.418",
+ "version": "0.1.419",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "clerum-mcp-host",
- "version": "0.1.418",
+ "version": "0.1.419",
"license": "SEE LICENSE IN LICENSE",
"dependencies": {
"@anthropic-ai/sdk": "^0.32.0",
diff --git a/mcp-host/package.json b/mcp-host/package.json
index 9d1730067..9a2138e22 100644
--- a/mcp-host/package.json
+++ b/mcp-host/package.json
@@ -1,6 +1,6 @@
{
"name": "clerum-mcp-host",
- "version": "0.1.418",
+ "version": "0.1.419",
"description": "MCP Host - reads Host CRD and provides LLM access via OpenAI/Claude",
"main": "dist/main.js",
"engines": {
diff --git a/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts b/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts
index b7e8a435d..bf35b72d0 100644
--- a/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts
+++ b/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts
@@ -280,6 +280,42 @@ describe('CodexLlmProxyClient', () => {
}
)
+ // T-MB-5 — the proxy answers `408 { error: 'request_timeout' }` when the
+ // body upload overruns its read deadline. The body never reached the
+ // upstream, so this is not an outage: it must not enter failover cooldown.
+ it('T-MB-5a classifies the proxy 408 request_timeout as a non-retryable ApiCallFailed', async () => {
+ const fetchFn = vi.fn(async () => Response.json({ error: 'request_timeout' }, { status: 408 }))
+ const client = new CodexLlmProxyClient({
+ runtimeUrl: resolveCodexProxyRuntimeUrl(
+ 'http://codex-llm-proxy.control-plane.svc.cluster.local:8080'
+ ),
+ readPlatformJwt: () => 'platform-jwt',
+ fetchFn: fetchFn as unknown as typeof fetch,
+ })
+ const err = await client
+ .stream({ executionTicket: 'ticket-123456', requestHash: 'a'.repeat(64), request: {} })
+ .then(
+ () => undefined,
+ (e: unknown) => e
+ )
+ expect(fetchFn).toHaveBeenCalledTimes(1)
+ expect(err).toBeInstanceOf(CodexProxyError)
+ expect(err).toMatchObject({
+ code: 'request_timeout',
+ message: 'proxy stream failed with 408 (request_timeout)',
+ })
+
+ const classified = new CodexSubscriptionProvider('gpt-5.3-codex', {} as never).classifyError(
+ err
+ )
+ expect(classified).toMatchObject({
+ code: LlmErrorCode.ApiCallFailed,
+ retryable: false,
+ providerCode: 'request_timeout',
+ })
+ expect(classifyFailoverClass(classified.code, classified.retryable)).toBeNull()
+ })
+
it('refuses a runtime URL that is not absolute', () => {
expect(
() =>
diff --git a/mcp-host/src/llm/__tests__/codexSubscription.test.ts b/mcp-host/src/llm/__tests__/codexSubscription.test.ts
index fe43cf967..7eb10745e 100644
--- a/mcp-host/src/llm/__tests__/codexSubscription.test.ts
+++ b/mcp-host/src/llm/__tests__/codexSubscription.test.ts
@@ -710,6 +710,23 @@ describe('CodexSubscriptionProvider', () => {
}
)
+ // T-MB-5 — the proxy's 408 for a body upload over its read deadline. The
+ // upstream never saw the body; a retryable class would cost the provider a
+ // failover cooldown for what is the Host's own slow upload.
+ it('T-MB-5c classifies request_timeout as a non-retryable ApiCallFailed', () => {
+ const provider = new CodexSubscriptionProvider('gpt-5.3-codex', deps() as never)
+ const message = 'proxy stream failed with 408 (request_timeout)'
+ const classified = provider.classifyError(new CodexProxyError('request_timeout', message))
+ expect(classified).toEqual({
+ code: LlmErrorCode.ApiCallFailed,
+ retryable: false,
+ message,
+ providerCode: 'request_timeout',
+ providerDispatched: true,
+ })
+ expect(classifyFailoverClass(classified.code, classified.retryable)).toBeNull()
+ })
+
it('keeps insufficient_scope distinguishable', () => {
const provider = new CodexSubscriptionProvider('gpt-5.3-codex', deps() as never)
const classified = provider.classifyError(
diff --git a/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts b/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts
index ec332f851..a4a189f16 100644
--- a/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts
+++ b/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts
@@ -255,6 +255,33 @@ describe('GrokLlmProxyClient', () => {
expect(fetchFn).toHaveBeenCalledTimes(1)
})
+ // T-MB-5 — the proxy answers `408 { error: 'request_timeout' }` when the
+ // body upload overruns its read deadline. The body never reached the
+ // upstream, so this is not an outage: it must not enter failover cooldown.
+ it('T-MB-5b classifies the proxy 408 request_timeout as a non-retryable ApiCallFailed', async () => {
+ const fetchFn = vi.fn(async () => Response.json({ error: 'request_timeout' }, { status: 408 }))
+ const err = await client(fetchFn)
+ .stream(STREAM_INPUT)
+ .then(
+ () => undefined,
+ (e: unknown) => e
+ )
+ expect(fetchFn).toHaveBeenCalledTimes(1)
+ expect(err).toBeInstanceOf(GrokProxyError)
+ expect(err).toMatchObject({
+ code: 'request_timeout',
+ message: 'proxy stream failed with 408 (request_timeout)',
+ })
+
+ const classified = new GrokSubscriptionProvider('grok-4.6', {} as never).classifyError(err)
+ expect(classified).toMatchObject({
+ code: LlmErrorCode.ApiCallFailed,
+ retryable: false,
+ providerCode: 'request_timeout',
+ })
+ expect(classifyFailoverClass(classified.code, classified.retryable)).toBeNull()
+ })
+
it('fails closed when the proxy emits an SSE error frame after headers', async () => {
const fetchFn = vi.fn().mockResolvedValue({
ok: true,
diff --git a/mcp-host/src/llm/__tests__/grokSubscription.test.ts b/mcp-host/src/llm/__tests__/grokSubscription.test.ts
index 0795d07f9..553dc2682 100644
--- a/mcp-host/src/llm/__tests__/grokSubscription.test.ts
+++ b/mcp-host/src/llm/__tests__/grokSubscription.test.ts
@@ -122,6 +122,23 @@ describe('GrokSubscriptionProvider', () => {
}
)
+ // T-MB-5 — the proxy's 408 for a body upload over its read deadline. The
+ // upstream never saw the body; a retryable class would cost the provider a
+ // failover cooldown for what is the Host's own slow upload.
+ it('T-MB-5d classifies request_timeout as a non-retryable ApiCallFailed', () => {
+ const provider = new GrokSubscriptionProvider('grok-4.6', deps() as never)
+ const message = 'proxy stream failed with 408 (request_timeout)'
+ const classified = provider.classifyError(new GrokProxyError('request_timeout', message))
+ expect(classified).toEqual({
+ code: LlmErrorCode.ApiCallFailed,
+ retryable: false,
+ message,
+ providerCode: 'request_timeout',
+ providerDispatched: true,
+ })
+ expect(classifyFailoverClass(classified.code, classified.retryable)).toBeNull()
+ })
+
it('still completes a successful terminal outcome', async () => {
const provider = new GrokSubscriptionProvider('grok-4.6', deps() as never)
await expect(