Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
test(mcp-host): require a 429 to read as rate_limited with its Retry-After
Alfredo Lopez8cccf9dc8b35
diff --git a/mcp-host/package-lock.json b/mcp-host/package-lock.jsonindex 2b1b9f464..69a172cf9 100644--- a/mcp-host/package-lock.json+++ b/mcp-host/package-lock.json@@ -1,12 +1,12 @@{"name": "clerum-mcp-host",- "version": "0.1.440",+ "version": "0.1.441","lockfileVersion": 3,"requires": true,"packages": {"": {"name": "clerum-mcp-host",- "version": "0.1.440",+ "version": "0.1.441","license": "SEE LICENSE IN LICENSE","dependencies": {"@anthropic-ai/sdk": "^0.32.0",diff --git a/mcp-host/package.json b/mcp-host/package.jsonindex 893808369..efdd06341 100644--- a/mcp-host/package.json+++ b/mcp-host/package.json@@ -1,6 +1,6 @@{"name": "clerum-mcp-host",- "version": "0.1.440",+ "version": "0.1.441","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.tsindex fd6369828..663023353 100644--- a/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts+++ b/mcp-host/src/llm/__tests__/codexLlmProxyClient.test.ts@@ -622,3 +622,77 @@ describe('CodexLlmProxyClient', () => {).rejects.toMatchObject({ code: 'rate_limited', dispatched: true })})})++// G1-6 (#720): a 429 is a rate limit, whether or not its body carries JSON,+// and its `Retry-After` (delta-seconds 1..3600) travels on the error.+describe('CodexLlmProxyClient rate limits', () => {+ const INPUT = {+ executionTicket: 'ticket-123456',+ requestHash: 'a'.repeat(64),+ request: {},+ }++ async function failure(response: Response) {+ const fetchFn = vi.fn<typeof fetch>(async () => response)+ const client = new CodexLlmProxyClient({+ runtimeUrl: 'http://proxy/completions',+ readPlatformJwt: () => 'platform-jwt',+ fetchFn,+ })+ const err = await client.stream(INPUT).then(+ () => undefined,+ (e: unknown) => e+ )+ return { err, fetchFn }+ }++ it('G1-6a reads a 429 with no JSON code as rate_limited with its Retry-After', async () => {+ const { err, fetchFn } = await failure(+ new Response('<html><body>Too Many Requests</body></html>', {+ status: 429,+ headers: { 'content-type': 'text/html', 'retry-after': '7' },+ })+ )+ // Liveness witness: the proxy hop ran and got the 429.+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toBeInstanceOf(CodexProxyError)+ expect(err).toMatchObject({ code: 'rate_limited', retryAfterMs: 7000 })+ })++ it('G1-6b carries the Retry-After of a JSON rate_limited reply', async () => {+ const { err, fetchFn } = await failure(+ Response.json({ error: 'rate_limited' }, { status: 429, headers: { 'retry-after': '2' } })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'rate_limited', retryAfterMs: 2000 })+ })++ it.each(['0', '3601', 'soon', '1.5', 'Wed, 21 Oct 2026 07:28:00 GMT', ''])(+ 'G1-6c drops the Retry-After value %j instead of guessing',+ async value => {+ const { err, fetchFn } = await failure(+ Response.json({ error: 'rate_limited' }, { status: 429, headers: { 'retry-after': value } })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'rate_limited' })+ expect((err as CodexProxyError).retryAfterMs).toBeUndefined()+ }+ )++ it('G1-6c keeps the JSON code a 429 carries', async () => {+ const { err } = await failure(Response.json({ error: 'budget_denied' }, { status: 429 }))+ expect(err).toMatchObject({ code: 'budget_denied' })+ })++ it('G1-6c keeps an HTML 502 as provider_unavailable with no Retry-After', async () => {+ const { err, fetchFn } = await failure(+ new Response('<html><body>502 Bad Gateway</body></html>', {+ status: 502,+ headers: { 'content-type': 'text/html', 'retry-after': '5' },+ })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'provider_unavailable' })+ expect((err as CodexProxyError).retryAfterMs).toBeUndefined()+ })+})diff --git a/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts b/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.tsindex a55719220..fda835189 100644--- a/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts+++ b/mcp-host/src/llm/__tests__/grokLlmProxyClient.test.ts@@ -531,3 +531,68 @@ describe('GrokLlmProxyClient', () => {expect(message).not.toMatch(/proxy stream failed/i)})})++// G1-6 (#720): a 429 is a rate limit, whether or not its body carries JSON,+// and its `Retry-After` (delta-seconds 1..3600) travels on the error.+describe('GrokLlmProxyClient rate limits', () => {+ async function failure(response: Response) {+ const fetchFn = vi.fn(async () => response)+ const err = await client(fetchFn)+ .stream(STREAM_INPUT)+ .then(+ () => undefined,+ (e: unknown) => e+ )+ return { err, fetchFn }+ }++ it('G1-6a reads a 429 with no JSON code as rate_limited with its Retry-After', async () => {+ const { err, fetchFn } = await failure(+ new Response('<html><body>Too Many Requests</body></html>', {+ status: 429,+ headers: { 'content-type': 'text/html', 'retry-after': '7' },+ })+ )+ // Liveness witness: the proxy hop ran and got the 429.+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toBeInstanceOf(GrokProxyError)+ expect(err).toMatchObject({ code: 'rate_limited', retryAfterMs: 7000 })+ })++ it('G1-6b carries the Retry-After of a JSON rate_limited reply', async () => {+ const { err, fetchFn } = await failure(+ Response.json({ error: 'rate_limited' }, { status: 429, headers: { 'retry-after': '2' } })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'rate_limited', retryAfterMs: 2000 })+ })++ it.each(['0', '3601', 'soon', '1.5', 'Wed, 21 Oct 2026 07:28:00 GMT', ''])(+ 'G1-6c drops the Retry-After value %j instead of guessing',+ async value => {+ const { err, fetchFn } = await failure(+ Response.json({ error: 'rate_limited' }, { status: 429, headers: { 'retry-after': value } })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'rate_limited' })+ expect((err as GrokProxyError).retryAfterMs).toBeUndefined()+ }+ )++ it('G1-6c keeps the JSON code a 429 carries', async () => {+ const { err } = await failure(Response.json({ error: 'budget_denied' }, { status: 429 }))+ expect(err).toMatchObject({ code: 'budget_denied' })+ })++ it('G1-6c keeps an HTML 502 as provider_unavailable with no Retry-After', async () => {+ const { err, fetchFn } = await failure(+ new Response('<html><body>502 Bad Gateway</body></html>', {+ status: 502,+ headers: { 'content-type': 'text/html', 'retry-after': '5' },+ })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'provider_unavailable' })+ expect((err as GrokProxyError).retryAfterMs).toBeUndefined()+ })+})diff --git a/mcp-host/src/llm/__tests__/providerAttemptAuthorizer.test.ts b/mcp-host/src/llm/__tests__/providerAttemptAuthorizer.test.tsindex 7d983fd49..b650be12f 100644--- a/mcp-host/src/llm/__tests__/providerAttemptAuthorizer.test.ts+++ b/mcp-host/src/llm/__tests__/providerAttemptAuthorizer.test.ts@@ -295,6 +295,23 @@ describe('ProviderAttemptAuthorizer', () => {})})+ // G1-6 (#720): a limiter in front of control-api can answer 429 with no JSON+ // code. It is a rate limit, not a provider outage.+ it('G1-6d reads an HTML 429 from the gateway as rate_limited', async () => {+ const { err, fetchFn } = await authorizeFailure(nginx(429, 'Too Many Requests'))+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toBeInstanceOf(CodexAuthorizeError)+ expect(err).toMatchObject({ code: 'rate_limited', message: 'authorize failed with 429' })+ })++ it('G1-6d keeps the JSON code a 429 carries', async () => {+ const { err, fetchFn } = await authorizeFailure(+ Response.json({ error: 'budget_denied' }, { status: 429 })+ )+ expect(fetchFn).toHaveBeenCalledTimes(1)+ expect(err).toMatchObject({ code: 'budget_denied' })+ })+it('T-R7-1c reads a 413 as payload_too_large whatever JSON code it carries', async () => {const { err, fetchFn } = await authorizeFailure(new Response(JSON.stringify({ error: 'invalid_request' }), {