Skip to content

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

TypeScript

fix(e2e): keep MCP probe within runtime budget

Alfredo Lopez98eb5ce1671a
diff --git a/scripts/e2e/e2e-wrc-hcc-context-noop-resync.sh b/scripts/e2e/e2e-wrc-hcc-context-noop-resync.sh
index 3bbb393e0..25c231a16 100755
--- a/scripts/e2e/e2e-wrc-hcc-context-noop-resync.sh
+++ b/scripts/e2e/e2e-wrc-hcc-context-noop-resync.sh
@@ -297,6 +297,12 @@ probe_mcp_tcp_connectivity() {
' "${MCP_NAME}.${MCP_NS}.svc.cluster.local" >/dev/null 2>&1
}
+mcp_data_plane_is_policy_blocked() {
+ resource_absent networkpolicy "$CONTEXT_POLICY" "$MCP_NS" &&
+ mcp_runtime_still_ready &&
+ ! probe_mcp_tcp_connectivity
+}
+
probe_mcp_business_signal() {
# shellcheck disable=SC2016 # JavaScript is intentionally literal.
kctl exec deployment/"$PROBE_NAME" -n "$HOST_NS" -c probe -- \
@@ -586,10 +592,10 @@ spec:
- name: probe
image: ${HCC_IMAGE}
imagePullPolicy: IfNotPresent
- command: [node, -e, "setInterval(()=>{},60000)"]
+ command: [sh, -c, "exec tail -f /dev/null"]
resources:
- requests: {cpu: 5m, memory: 16Mi}
- limits: {cpu: 50m, memory: 48Mi}
+ requests: {cpu: 5m, memory: 24Mi}
+ limits: {cpu: 50m, memory: 128Mi}
securityContext:
allowPrivilegeEscalation: false
capabilities: {drop: [ALL]}
@@ -671,13 +677,9 @@ ORIGINAL_POLICY_SNAPSHOT="$(context_policy_snapshot)" ||
kctl delete networkpolicy "$CONTEXT_POLICY" -n "$MCP_NS" --wait=true --timeout=60s >/dev/null
resource_absent networkpolicy "$CONTEXT_POLICY" "$MCP_NS" ||
die "fault injection did not remove the Context ingress policy"
-if probe_mcp_tcp_connectivity; then
- die "MCP data plane remained reachable after deleting its only Context ingress policy"
-fi
-resource_absent networkpolicy "$CONTEXT_POLICY" "$MCP_NS" ||
- die "Context ingress policy was recreated before the isolation failure could be attributed"
-mcp_runtime_still_ready ||
- die "MCP workload or Service endpoints became unhealthy during the policy fault"
+wait_until 30 "Calico to enforce the missing Context ingress policy while MCP stays Ready" \
+ mcp_data_plane_is_policy_blocked ||
+ die "MCP did not become policy-isolated while its workload and Service endpoints stayed Ready"
ok "policy deletion produced a real fail-closed MCP data-plane interruption"
wait_until "$RESYNC_TIMEOUT_SECONDS" \
diff --git a/scripts/tests/test-wrc-hcc-contracts-gate.sh b/scripts/tests/test-wrc-hcc-contracts-gate.sh
index 43555a67b..b01ccb524 100755
--- a/scripts/tests/test-wrc-hcc-contracts-gate.sh
+++ b/scripts/tests/test-wrc-hcc-contracts-gate.sh
@@ -89,6 +89,8 @@ if grep -Fq 'CONTEXT_MAPPER_NETPOL_RESYNC_SEC=${RESYNC_SECONDS}' "$RUNTIME_GATE"
grep -Fq 'kctl delete networkpolicy "$CONTEXT_POLICY"' "$RUNTIME_GATE" &&
grep -Fq 'probe_mcp_tcp_connectivity' "$RUNTIME_GATE" &&
grep -Fq 'mcp_runtime_still_ready' "$RUNTIME_GATE" &&
+ grep -Fq 'mcp_data_plane_is_policy_blocked' "$RUNTIME_GATE" &&
+ grep -Fq 'limits: {cpu: 50m, memory: 128Mi}' "$RUNTIME_GATE" &&
grep -Fq 'context_policy_recreated_from_snapshot' "$RUNTIME_GATE" &&
grep -Fq 'hcc_identity_is_stable' "$RUNTIME_GATE" &&
grep -Fq 'restore_hcc_resync_config' "$RUNTIME_GATE"; then