Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
test(llm-proxy): budget the finalize retry in the shutdown grace
Alfredo Lopez1a097f22abf5
diff --git a/codex-llm-proxy/test/deployManifest.test.ts b/codex-llm-proxy/test/deployManifest.test.tsindex 32d06d061..54441568b 100644--- a/codex-llm-proxy/test/deployManifest.test.ts+++ b/codex-llm-proxy/test/deployManifest.test.ts@@ -62,13 +62,14 @@ describe('codex-llm-proxy base manifest', () => {const streamMs = Math.min(configuredStreamMs, STREAM_LIMITS.maxStreamDurationMs)// After SIGTERM the server stops accepting connections, but a request that// already arrived can still wait in the queue, read its body, redeem its- // ticket, stream to the cap and finalize.+ // ticket, stream to the cap and finalize. finalizeQuietly retries a failed+ // finalize once, so finalize is budgeted as two control-api calls.const worstCaseMs =STREAM_LIMITS.maxQueueWaitMs +BODY_READ_DEADLINE_MS +CONTROL_API_REQUEST_TIMEOUT_MS + // redeemstreamMs +- CONTROL_API_REQUEST_TIMEOUT_MS // finalize+ 2 * CONTROL_API_REQUEST_TIMEOUT_MS // finalize and its one retryconst grace = Number(scalar(MANIFEST, 'terminationGracePeriodSeconds'))expect(grace).toBe(Math.ceil(worstCaseMs / 1000) + SHUTDOWN_MARGIN_SECONDS)})diff --git a/grok-llm-proxy/test/deployManifest.test.ts b/grok-llm-proxy/test/deployManifest.test.tsindex 40ded82f8..f68cf22cc 100644--- a/grok-llm-proxy/test/deployManifest.test.ts+++ b/grok-llm-proxy/test/deployManifest.test.ts@@ -67,13 +67,14 @@ describe('grok-llm-proxy base manifest', () => {const streamMs = Math.min(configuredStreamMs, STREAM_LIMITS.maxStreamDurationMs)// After SIGTERM the server stops accepting connections, but a request that// already arrived can still wait in the queue, read its body, redeem its- // ticket, stream to the cap and finalize.+ // ticket, stream to the cap and finalize. finalizeQuietly retries a failed+ // finalize once, so finalize is budgeted as two control-api calls.const worstCaseMs =STREAM_LIMITS.maxQueueWaitMs +BODY_READ_DEADLINE_MS +CONTROL_API_REQUEST_TIMEOUT_MS + // redeemstreamMs +- CONTROL_API_REQUEST_TIMEOUT_MS // finalize+ 2 * CONTROL_API_REQUEST_TIMEOUT_MS // finalize and its one retryconst grace = Number(scalar(MANIFEST, 'terminationGracePeriodSeconds'))expect(grace).toBe(Math.ceil(worstCaseMs / 1000) + SHUTDOWN_MARGIN_SECONDS)})