Corprun

Makefile

@@ -40,6 +40,7 @@ container:
 		$(if $(filter docker,$(CONTAINER_ENGINE)),--net=host) \
 		-v ~/.config/gcloud:/home/evalbench/.config/gcloud \
 		-e GOOGLE_CLOUD_PROJECT=cloud-db-nl2sql \
+		-e MESOP_XSRF_CHECK=false \
 		--cap-add=SYS_PTRACE	\
 		-p 3000:3000 \
 		-p 50051:50051 \
@@ -70,6 +71,10 @@ push:
 	$(CONTAINER_ENGINE) image tag evalbench:latest us-central1-docker.pkg.dev/cloud-db-nl2sql/evalbench/eval_server:latest
 	$(CONTAINER_ENGINE) push us-central1-docker.pkg.dev/cloud-db-nl2sql/evalbench/eval_server:latest
 
+push-corprun:
+	$(CONTAINER_ENGINE) image tag evalbench:latest us-central1-docker.pkg.dev/evalbench-dev/cr-images/eval_server:latest
+	$(CONTAINER_ENGINE) push us-central1-docker.pkg.dev/evalbench-dev/cr-images/eval_server:latest
+
 deploy:
 	gcloud container clusters get-credentials evalbench-directpath-cluster --zone us-central1-c --project cloud-db-nl2sql
 	kubectl apply -f evalbench_service/k8s/namespace.yaml
@@ -88,6 +93,22 @@ deploy-test:
 	kubectl apply -f evalbench_service/k8s/evalbench-test.yaml
 	kubectl apply -f evalbench_service/k8s/vertical-autoscale-test.yaml
 
+deploy-corprun:
+	gcloud run deploy evalbench \
+		--project=evalbench-dev \
+		--region=us-central1 \
+		--image=us-central1-docker.pkg.dev/evalbench-dev/cr-images/eval_server:latest \
+		--port=3000 \
+		--memory=2Gi \
+		--service-account=crsvc-evalbench@evalbench-dev.iam.gserviceaccount.com \
+		--set-env-vars CLOUD_RUN=True,GOOGLE_CLOUD_PROJECT=evalbench-dev,MESOP_XSRF_CHECK=false \
+		--ingress=internal-and-cloud-load-balancing \
+		--network=cr-infra-vpc-network \
+		--subnet=cr-infra-subnetwork \
+		--vpc-egress=all-traffic \
+		--add-volume=name=session-files,type=cloud-storage,bucket=evalbench-sessions-cloud-db-nl2sql \
+		--add-volume-mount=volume=session-files,mount-path=/tmp_session_files
+
 undeploy:
 	gcloud container clusters get-credentials evalbench-directpath-cluster --zone us-central1-c --project cloud-db-nl2sql
 	kubectl delete -f evalbench_service/k8s/evalbench.yaml

cloudbuild.yaml

@@ -3,11 +3,13 @@ steps:
 - name: 'gcr.io/cloud-builders/docker'
   args: ['build', '-t', 'us-central1-docker.pkg.dev/cloud-db-nl2sql/evalbench/eval_server:$COMMIT_SHA', '-f', 'evalbench_service/Dockerfile', '.']
 - name: 'us-central1-docker.pkg.dev/cloud-db-nl2sql/evalbench/eval_server:$COMMIT_SHA'
-  args: ['evalbench/run.sh']
+  dir: '/evalbench'
+  args: ['/evalbench/evalbench/run.sh']
   env:
   - 'EVAL_GCP_PROJECT_ID=${_VAR_PROJECT}'
   - 'EVAL_GCP_PROJECT_REGION=${_VAR_REGION}'
   - 'EVAL_CONFIG=${_VAR_EVAL_CONFIG}'
+  - 'UV_CACHE_DIR=/tmp/uv-cache'
 images:
 - 'us-central1-docker.pkg.dev/cloud-db-nl2sql/evalbench/eval_server:$COMMIT_SHA'
 substitutions:

evalbench_service/Dockerfile

@@ -27,13 +27,19 @@ RUN mkdir -p /home/evalbench && \
 
 COPY . evalbench
 WORKDIR evalbench
-RUN uv sync
+RUN uv pip install --system --break-system-packages -r requirements.txt
+RUN uv pip install --system --break-system-packages ./viewer
+RUN uv pip install --system --break-system-packages .
 
 RUN ln -s /usr/bin/python3 /usr/bin/python
 RUN make proto -f ./Makefile
 RUN mkdir /tmp_session_files /tmp_sessions
 RUN cp /evalbench/evalbench_service/supervisord.conf /evalbench/supervisord.conf
+RUN cp /evalbench/evalbench_service/entrypoint.sh /evalbench/entrypoint.sh && chmod +x /evalbench/entrypoint.sh
+RUN chmod +x /evalbench/evalbench/run.sh
 RUN chown -R 65532:65532 /evalbench /tmp /tmp_session_files /tmp_sessions /home/evalbench
 
-CMD ["/usr/bin/supervisord", "-c", "/evalbench/supervisord.conf"]
+USER 65532
+WORKDIR /evalbench
+CMD ["/evalbench/entrypoint.sh"]
 EXPOSE 50051 3000

evalbench_service/entrypoint.sh

@@ -0,0 +1,12 @@
+#!/bin/bash
+# evalbench_service/entrypoint.sh
+
+if [[ "$CLOUD_RUN" == "True" ]]; then
+    echo "Cloud Run detected. Starting only gunicorn frontend on port ${PORT:-3000}..."
+    # Ensure we are in the viewer directory for gunicorn to find main:me
+    cd /evalbench/viewer
+    exec gunicorn -w 4 -k gevent main:me --bind :${PORT:-3000} --forwarded-allow-ips="*"
+else
+    echo "Starting supervisord to manage multiple processes..."
+    exec /usr/bin/supervisord -c /evalbench/supervisord.conf
+fi

evalbench_service/supervisord.conf

@@ -6,7 +6,7 @@ logfile_backups=5
 
 
 [program:evalbench_server]
-command=uv run evalbench/eval_server.py
+command=python evalbench/eval_server.py
 directory=/evalbench
 autostart=true
 autorestart=true
@@ -16,7 +16,7 @@ stderr_logfile=/dev/stderr
 stderr_logfile_maxbytes=0
 
 [program:evalbench_frontend]
-command=uv run gunicorn -w 4 -k gevent main:me --bind :3000
+command=gunicorn -w 4 -k gevent main:me --bind :3000 --forwarded-allow-ips="*"
 directory=/evalbench/viewer
 autostart=true
 autorestart=true

viewer/main.py

@@ -6,6 +6,15 @@
 
 logging.basicConfig(level=logging.INFO)
 
+# Manually enable debug mode to bypass XSRF check if needed
+# (e.g. when running in container behind a proxy)
+if os.environ.get("MESOP_XSRF_CHECK") == "false":
+    try:
+        import mesop.runtime as mesop_runtime
+        mesop_runtime.enable_debug_mode()
+    except Exception as e:
+        logging.error(f"Failed to enable debug mode: {e}")
+
 try:
     import dashboard
     import conversations
@@ -54,37 +63,54 @@ class State:
     conversation_index: int = 0
 
 
-@me.page(
-    path="/",
-    title="Evalbench",
-    stylesheets=[
-        "data:",
-        "data:text/css;charset=utf-8,"
-        ".mdc-tooltip__surface%20%7B%0A"
-        "%20%20max-height%3A%20none%20%21important%3B%0A"
-        "%20%20max-width%3A%20none%20%21important%3B%0A"
-        "%20%20white-space%3A%20pre-wrap%20%21important%3B%0A"
-        "%7D",
-    ],
-)
-def app():
-    state = me.state(State)
-
+def get_results_dir():
     # Check multiple locations for results directory
     results_dir_candidates = [
         "/tmp_session_files/results",
         os.path.join(os.path.dirname(os.path.dirname(__file__)), "results"),
         os.path.join(os.getcwd(), "results"),
     ]
 
-    results_dir = None
     for candidate in results_dir_candidates:
         if os.path.exists(candidate) and os.path.isdir(candidate):
-            results_dir = candidate
-            break
+            return candidate
+
+    return results_dir_candidates[1]  # Fallback to default
+
+
+def on_load(e: me.LoadEvent):
+    state = me.state(State)
+    results_dir = get_results_dir()
+    directories = []
+    if os.path.exists(results_dir):
+        # List directories only
+        directories = [
+            d
+            for d in os.listdir(results_dir)
+            if os.path.isdir(os.path.join(results_dir, d))
+        ]
+
+    job_id = me.query_params.get("job_id") or me.query_params.get("jobid")
+    if job_id and job_id in directories:
+        state.selected_directory = job_id
 
-    if results_dir is None:
-        results_dir = results_dir_candidates[1]  # Fallback to default
+
+@me.page(
+    path="/",
+    title="Evalbench",
+    on_load=on_load,
+    security_policy=me.SecurityPolicy(
+        dangerously_disable_trusted_types=True,
+        cross_origin_opener_policy="same-origin",
+    ),
+    stylesheets=[
+        "data:",
+        "/static/custom.css",
+    ],
+)
+def app():
+    state = me.state(State)
+    results_dir = get_results_dir()
 
     directories = []
     if os.path.exists(results_dir):

viewer/static/custom.css

@@ -0,0 +1,5 @@
+.mdc-tooltip__surface {
+  max-height: none !important;
+  max-width: none !important;
+  white-space: pre-wrap !important;
+}