Skip to content

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

TypeScript

test(llm-proxy): require a 768Mi memory request in both proxies

Alfredo Lopezef45b7a1fca5
diff --git a/codex-llm-proxy/test/deployManifest.test.ts b/codex-llm-proxy/test/deployManifest.test.ts
index 54441568b..6c5565088 100644
--- a/codex-llm-proxy/test/deployManifest.test.ts
+++ b/codex-llm-proxy/test/deployManifest.test.ts
@@ -4,8 +4,8 @@ import { CONTROL_API_REQUEST_TIMEOUT_MS } from '../src/controlApiClient.js'
import { BODY_READ_DEADLINE_MS, STREAM_LIMITS } from '../src/requestLimits.js'
// The base manifest must agree with the limits this process enforces. Each
-// check is a relation with a code constant or a recorded measurement, so a
-// change on either side that breaks the relation fails here.
+// check is a relation with a code constant, a recorded measurement or a
+// recorded owner decision, so a change on either side fails here.
const MANIFEST = readFileSync(
new URL('../../deploy/base/control-plane/codex-llm-proxy.yaml', import.meta.url),
@@ -24,6 +24,11 @@ const SHUTDOWN_MARGIN_SECONDS = 20
*/
const D5_CAPPED_PEAK_RSS_MIB = 790.3
const MEMORY_HEADROOM = 1.25
+/**
+ * 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 = 768
function activeLines(yaml: string): string[] {
return yaml.split('\n').filter((line) => !line.trimStart().startsWith('#'))
@@ -74,7 +79,7 @@ describe('codex-llm-proxy base manifest', () => {
expect(grace).toBe(Math.ceil(worstCaseMs / 1000) + SHUTDOWN_MARGIN_SECONDS)
})
- it('T-DEP-2 caps the heap below the memory limit and keeps the limit above the D5 peak', () => {
+ it('T-DEP-2 caps the heap below the memory limit, keeps the limit above the D5 peak and requests 768Mi', () => {
const nodeOptions = activeLines(MANIFEST).findIndex((line) => /name:\s*NODE_OPTIONS\s*$/.test(line))
expect(nodeOptions, 'the container must set NODE_OPTIONS').toBeGreaterThanOrEqual(0)
const heap = /--max-old-space-size=(\d+)/.exec(activeLines(MANIFEST)[nodeOptions + 1] ?? '')
@@ -82,6 +87,7 @@ describe('codex-llm-proxy base manifest', () => {
const limit = resourceMemory('limits')
expect(Number(heap![1])).toBeLessThan(limit)
expect(resourceMemory('requests')).toBeLessThanOrEqual(limit)
+ expect(resourceMemory('requests')).toBe(MEMORY_REQUEST_MIB)
expect(limit).toBeGreaterThanOrEqual(Math.ceil(D5_CAPPED_PEAK_RSS_MIB * MEMORY_HEADROOM))
})
diff --git a/grok-llm-proxy/test/deployManifest.test.ts b/grok-llm-proxy/test/deployManifest.test.ts
index f68cf22cc..fb8a3f3b6 100644
--- a/grok-llm-proxy/test/deployManifest.test.ts
+++ b/grok-llm-proxy/test/deployManifest.test.ts
@@ -4,8 +4,8 @@ import { CONTROL_API_REQUEST_TIMEOUT_MS } from '../src/controlApiClient.js'
import { BODY_READ_DEADLINE_MS, STREAM_LIMITS } from '../src/requestLimits.js'
// The base manifest must agree with the limits this process enforces. Each
-// check is a relation with a code constant or a recorded measurement, so a
-// change on either side that breaks the relation fails here.
+// check is a relation with a code constant, a recorded measurement or a
+// recorded owner decision, so a change on either side fails here.
const MANIFEST = readFileSync(
new URL('../../deploy/base/control-plane/grok-llm-proxy.yaml', import.meta.url),
@@ -29,6 +29,12 @@ const SHUTDOWN_MARGIN_SECONDS = 20
*/
const D5_CAPPED_PEAK_RSS_MIB = 509.8
const 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.
+ */
+const MEMORY_REQUEST_MIB = 768
function activeLines(yaml: string): string[] {
return yaml.split('\n').filter((line) => !line.trimStart().startsWith('#'))
@@ -79,7 +85,7 @@ describe('grok-llm-proxy base manifest', () => {
expect(grace).toBe(Math.ceil(worstCaseMs / 1000) + SHUTDOWN_MARGIN_SECONDS)
})
- it('T-DEP-2 caps the heap below the memory limit and keeps the limit above the D5 peak', () => {
+ it('T-DEP-2 caps the heap below the memory limit, keeps the limit above the D5 peak and requests 768Mi', () => {
const nodeOptions = activeLines(MANIFEST).findIndex((line) => /name:\s*NODE_OPTIONS\s*$/.test(line))
expect(nodeOptions, 'the container must set NODE_OPTIONS').toBeGreaterThanOrEqual(0)
const heap = /--max-old-space-size=(\d+)/.exec(activeLines(MANIFEST)[nodeOptions + 1] ?? '')
@@ -87,6 +93,7 @@ describe('grok-llm-proxy base manifest', () => {
const limit = resourceMemory('limits')
expect(Number(heap![1])).toBeLessThan(limit)
expect(resourceMemory('requests')).toBeLessThanOrEqual(limit)
+ expect(resourceMemory('requests')).toBe(MEMORY_REQUEST_MIB)
expect(limit).toBeGreaterThanOrEqual(Math.ceil(D5_CAPPED_PEAK_RSS_MIB * MEMORY_HEADROOM))
})
diff --git a/scripts/tests/test-codex-llm-proxy-deploy-contract.sh b/scripts/tests/test-codex-llm-proxy-deploy-contract.sh
index de355dd12..6e46fc4f6 100755
--- a/scripts/tests/test-codex-llm-proxy-deploy-contract.sh
+++ b/scripts/tests/test-codex-llm-proxy-deploy-contract.sh
@@ -100,7 +100,9 @@ if named_containers != ["codex-llm-proxy"]:
# Measured on the shipping design (#739 D5): eight 8 MiB streams, two 24 MiB
# visual streams and three queued 8 MiB bodies peaked at 790 MiB of RSS with a
# 384 MiB old space and at 886-1009 MiB without it. The capped peak does not fit
-# 768Mi; the limit is that peak plus 25 %, rounded up to 1Gi.
+# 768Mi; the limit is that peak plus 25 %, rounded up to 1Gi. 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(
@@ -108,8 +110,8 @@ heap_cap = re.search(
)
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) != "256Mi":
- errors.append("proxy memory request must be 256Mi")
+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":
errors.append("proxy must cap the V8 old space at 384 MiB through NODE_OPTIONS")
diff --git a/scripts/tests/test-grok-llm-proxy-deploy-contract.sh b/scripts/tests/test-grok-llm-proxy-deploy-contract.sh
index f9c617d81..e92e3f21c 100755
--- a/scripts/tests/test-grok-llm-proxy-deploy-contract.sh
+++ b/scripts/tests/test-grok-llm-proxy-deploy-contract.sh
@@ -88,7 +88,9 @@ if "CONTROL_API_INTERNAL_SERVICE_TOKENS" in text:
# 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).
+# 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
+# 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(
@@ -96,8 +98,8 @@ heap_cap = re.search(
)
if not memory_limit or memory_limit.group(1) != "768Mi":
errors.append("proxy memory limit must be 768Mi")
-if not memory_request or memory_request.group(1) != "256Mi":
- errors.append("proxy memory request must be 256Mi")
+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":
errors.append("proxy must cap the V8 old space at 384 MiB through NODE_OPTIONS")