fix: drop MinIO tmpfs cap for OOM proof tests

.github/workflows/helm-install-test.yml

@@ -107,6 +107,7 @@ jobs:
             --set dashboard.frontend.image.tag=latest \
             --set dashboard.frontend.image.pullPolicy=Always \
             --set frontproxy.enabled=true \
+            --set frontproxy.metrics.serviceMonitor.enabled=false \
             --set replicaCount=3 \
             --set resources.limits.cpu=100m \
             --set resources.requests.cpu=50m \

Makefile

@@ -101,14 +101,15 @@ test-run:
 		exit $$EXIT_CODE
 
 # OOM proof test: runs s3proxy in a 256MB container and hammers it
+OOM_COMPOSE = -f tests/docker-compose.yml -f tests/docker-compose.oom.yml --profile oom
 test-oom:
-	@docker compose -f tests/docker-compose.yml --profile oom down 2>/dev/null || true
-	@docker compose -f tests/docker-compose.yml --profile oom up -d --build
+	@docker compose $(OOM_COMPOSE) down 2>/dev/null || true
+	@docker compose $(OOM_COMPOSE) up -d --build
 	@sleep 5
 	@AWS_ACCESS_KEY_ID=minioadmin AWS_SECRET_ACCESS_KEY=minioadmin \
 		uv run pytest -v tests/integration/test_memory_leak.py; \
 		EXIT_CODE=$$?; \
-		docker compose -f tests/docker-compose.yml --profile oom down; \
+		docker compose $(OOM_COMPOSE) down; \
 		exit $$EXIT_CODE
 
 # E2E cluster commands

tests/docker-compose.oom.yml

@@ -0,0 +1,6 @@
+# OOM tests upload 5GB+ into MinIO (e.g. 20 concurrent 256MB PUTs).
+# The default compose caps MinIO at 4g tmpfs for integration shards; drop
+# that cap here so MinIO uses disk-backed storage like before #117.
+services:
+  minio:
+    tmpfs: !reset null