Evenfire is an intelligence layer platform for running LLM agents that take real actions, on your infrastructure
TypeScript
test(grok-llm-proxy): size the memory limit for the visual stream slot
Alfredo Lopezf948a9dd311c
diff --git a/grok-llm-proxy/test/deployManifest.test.ts b/grok-llm-proxy/test/deployManifest.test.tsindex fb8a3f3b6..0ff4f0a43 100644--- a/grok-llm-proxy/test/deployManifest.test.ts+++ b/grok-llm-proxy/test/deployManifest.test.ts@@ -24,15 +24,16 @@ const WORKFLOW_RECIPES = readFileSync(/** Headroom between the last in-flight request finishing and SIGKILL. */const SHUTDOWN_MARGIN_SECONDS = 20/**- * D5 (#739): peak RSS with eight 8 MiB streams and three queued 8 MiB bodies,- * heap capped at 384 MiB (tsc build, one process).+ * D5 (#739) plus the visual slot (#784): peak RSS with eight 8 MiB streams,+ * one ~36 MB V2 stream (a 20 MiB PNG and 8 MiB of text) and three queued+ * 8 MiB bodies, heap capped at 384 MiB (tsc build, one process, upstream+ * request through undici). D5 alone measured 511 MiB (#739: 509.8).*/-const D5_CAPPED_PEAK_RSS_MIB = 509.8+const D5_CAPPED_PEAK_RSS_MIB = 775.4const MEMORY_HEADROOM = 1.25/**- * Owner decision on review M4 (#739): the request equals the limit, so the- * pod's memory use can never exceed its request and a busy node does not- * schedule it on memory it cannot give it under load.+ * Owner decision on review M4 (#739): the request sits near the D5 peak, so a+ * busy node does not schedule the pod on memory it cannot give it under load.*/const MEMORY_REQUEST_MIB = 768diff --git a/scripts/tests/test-grok-llm-proxy-deploy-contract.sh b/scripts/tests/test-grok-llm-proxy-deploy-contract.shindex e92e3f21c..e80a0d7f5 100755--- a/scripts/tests/test-grok-llm-proxy-deploy-contract.sh+++ b/scripts/tests/test-grok-llm-proxy-deploy-contract.sh@@ -86,18 +86,18 @@ if "CONTROL_API_INTERNAL_SERVICE_TOKENS" in text:errors.append("proxy must not receive the full token map")# Eight 8 MiB streams and three queued 8 MiB bodies (#739 D5) peaked at 510 MiB-# of RSS with a 384 MiB old space and at 480-511 MiB with an uncapped heap. That-# is past the former 256Mi limit and under 768Mi, so the pod gets a 768Mi limit-# and the same heap cap as codex-llm-proxy (whose visual slots need 1Gi). The-# request equals the limit (owner decision on review M4), so a busy node does not+# of RSS with a 384 MiB old space. One ~36 MB V2 stream in the visual slot+# (#784) on top of that load peaked at 775 MiB, past 768Mi; the limit is that+# peak plus 25 %, rounded up to 1Gi, as on codex-llm-proxy. The request is+# 768Mi (owner decision on review M4), near the peak, so a busy node does not# schedule the pod on memory it cannot give it under load.memory_limit = re.search(r"limits:\n\s+cpu: \S+\n\s+memory: (\S+)", text)memory_request = re.search(r"requests:\n\s+cpu: \S+\n\s+memory: (\S+)", text)heap_cap = re.search(r"- name: NODE_OPTIONS\n\s+value: \"--max-old-space-size=(\d+)\"", text)-if not memory_limit or memory_limit.group(1) != "768Mi":- errors.append("proxy memory limit must be 768Mi")+if not memory_limit or memory_limit.group(1) != "1Gi":+ errors.append("proxy memory limit must be 1Gi")if not memory_request or memory_request.group(1) != "768Mi":errors.append("proxy memory request must be 768Mi")if not heap_cap or heap_cap.group(1) != "384":