TensorFlow Serving

This document describes how to configure your Google Kubernetes Engine deployment so that you can use Google Cloud Managed Service for Prometheus to collect metrics from TensorFlow Serving. This document shows you how to do the following:

Set up TF Serving to report metrics.
Access a dashboard in Cloud Monitoring to view the metrics.

These instructions apply only if you are using managed collection with Managed Service for Prometheus. If you are using self-deployed collection, then see the TF Serving documentation for installation information.

These instructions are provided as an example and are expected to work in most Kubernetes environments. If you are having trouble installing an application or exporter due to restrictive security or organizational policies, then we recommend you consult open-source documentation for support.

For information about TensorFlow Serving, see TF Serving . For information about setting up TF Serving on Google Kubernetes Engine, see the GKE guide for TF Serving .

Prerequisites

To collect metrics from TF Serving by using Managed Service for Prometheus and managed collection, your deployment must meet the following requirements:

Your cluster must be running Google Kubernetes Engine version 1.28.15-gke.2475000 or later.
You must be running Managed Service for Prometheus with managed collection enabled. For more information, see Get started with managed collection .

TF Serving exposes Prometheus-format metrics when the --monitoring_config_file flag is used to specify a file containing a MonitoringConfig protocol buffer.

The following is an example of a MonitoringConfig protocol buffer:

 prometheus_config {
  enable: true,
  path: "/monitoring/prometheus/metrics"
}

If you are following the Google Kubernetes Engine set-up guide, Serve a model with a single GPU in GKE , then the MonitoringConfig protocol buffer is defined as part of the default setup.

If you are setting up TF Serving yourself, then do the following to specify the MonitoringConfig protocol buffer:

Create a file named monitoring_config.txt containing the MonitoringConfig protocol buffer in the model directory, before uploading the directory to the Cloud Storage bucket.

Upload the model directory to the Cloud Storage bucket:

gcloud storage cp MODEL_DIRECTORY 
gs:// CLOUD_STORAGE_BUCKET_NAME 
--recursive

Set the environment variable PATH_TO_MONITORING_CONFIG to the path of the uploaded monitoring_config.txt file, for example:
```
export PATH_TO_MONITORING_CONFIG=/data/tfserve-model-repository/monitoring_config.txt
```

Add the following flag and value to the container's command in your container's deployment YAML file:

"--monitoring_config=$PATH_TO_MONITORING_CONFIG"

For example, your command might look like the following:

command: [ "tensorflow_model_server", "--model_name=$MODEL_NAME", "--model_base_path=/data/tfserve-model-repository/$MODEL_NAME", "--rest_api_port=8000", "--monitoring_config_file=$PATH_TO_MONITORING_CONFIG" ]

Modify the TF Serving configuration

Modify the TF Serving configuration as shown in the following example:

  # 
  
 Copyright 
  
 2025 
  
 Google 
  
 LLC 
 # 
 # 
  
 Licensed 
  
 under 
  
 the 
  
 Apache 
  
 License 
 , 
  
 Version 
  
 2.0 
  
 ( 
 the 
  
 "License" 
 ); 
 # 
  
 you 
  
 may 
  
 not 
  
 use 
  
 this 
  
 file 
  
 except 
  
 in 
  
 compliance 
  
 with 
  
 the 
  
 License 
 . 
 # 
  
 You 
  
 may 
  
 obtain 
  
 a 
  
 copy 
  
 of 
  
 the 
  
 License 
  
 at 
 # 
 # 
  
 https 
 : 
 //www.apache.org/licenses/LICENSE-2.0 
 # 
 # 
  
 Unless 
  
 required 
  
 by 
  
 applicable 
  
 law 
  
 or 
  
 agreed 
  
 to 
  
 in 
  
 writing 
 , 
  
 software 
 # 
  
 distributed 
  
 under 
  
 the 
  
 License 
  
 is 
  
 distributed 
  
 on 
  
 an 
  
 "AS IS" 
  
 BASIS 
 , 
 # 
  
 WITHOUT 
  
 WARRANTIES 
  
 OR 
  
 CONDITIONS 
  
 OF 
  
 ANY 
  
 KIND 
 , 
  
 either 
  
 express 
  
 or 
  
 implied 
 . 
 # 
  
 See 
  
 the 
  
 License 
  
 for 
  
 the 
  
 specific 
  
 language 
  
 governing 
  
 permissions 
  
 and 
 # 
  
 limitations 
  
 under 
  
 the 
  
 License 
 . 
 apiVersion 
 : 
  
 apps 
 / 
 v1 
 kind 
 : 
  
 Deployment 
 metadata 
 : 
  
 name 
 : 
  
 tfserve 
 - 
 deployment 
  
 labels 
 : 
  
 app 
 : 
  
 tfserve 
 - 
 server 
 spec 
 : 
  
 selector 
 : 
  
 matchLabels 
 : 
  
 app 
 : 
  
 tfserve 
  
 replicas 
 : 
  
 1 
  
 template 
 : 
  
 metadata 
 : 
  
 labels 
 : 
  
 app 
 : 
  
 tfserve 
  
 annotations 
 : 
  
 gke 
 - 
 gcsfuse 
 / 
 volumes 
 : 
  
 ' 
 true 
 ' 
  
 spec 
 : 
  
 nodeSelector 
 : 
  
 cloud 
 . 
 google 
 . 
 com 
 / 
 gke 
 - 
 accelerator 
 : 
  
 nvidia 
 - 
 l4 
  
 containers 
 : 
  
 - 
  
 name 
 : 
  
 tfserve 
 - 
 server 
  
 image 
 : 
  
 ' 
 tensorflow 
 / 
 serving 
 : 
 2.13.1 
 - 
 gpu 
 ' 
  
 command 
 : 
  
 - 
  
 tensorflow_model_server 
  
 - 
  
 ' 
 -- 
 model_name 
 = 
 $ 
 MODEL_NAME 
 ' 
  
 - 
  
 ' 
 -- 
 model_base_path 
 = 
 / 
 data 
 / 
 tfserve 
 - 
 model 
 - 
 repository 
 / 
 $ 
 MODEL_NAME 
 ' 
  
 - 
  
 ' 
 -- 
 rest_api_port 
 = 
 8000 
 ' 
 + 
  
 - 
  
 ' 
 -- 
 monitoring_config_file 
 = 
 $ 
 PATH_TO_MONITORING_CONFIG 
 ' 
  
 ports 
 : 
  
 - 
  
 name 
 : 
  
 http 
  
 containerPort 
 : 
  
 8000 
  
 - 
  
 name 
 : 
  
 grpc 
  
 containerPort 
 : 
  
 8500 
  
 resources 
 : 
  
 ... 
  
 volumeMounts 
 : 
  
 - 
  
 name 
 : 
  
 gcs 
 - 
 fuse 
 - 
 csi 
 - 
 vol 
  
 mountPath 
 : 
  
 / 
 data 
  
 readOnly 
 : 
  
 false 
  
 serviceAccountName 
 : 
  
 $ 
 K8S_SA_NAME 
  
 volumes 
 : 
  
 - 
  
 name 
 : 
  
 gcs 
 - 
 fuse 
 - 
 csi 
 - 
 vol 
  
 csi 
 : 
  
 driver 
 : 
  
 gcsfuse 
 . 
 csi 
 . 
 storage 
 . 
 gke 
 . 
 io 
  
 readOnly 
 : 
  
 false 
  
 volumeAttributes 
 : 
  
 bucketName 
 : 
  
 $ 
 GSBUCKET 
  
 mountOptions 
 : 
  
 implicit 
 - 
 dirs

You must add any lines preceded by the + symbol to your configuration.

To apply configuration changes from a local file, run the following command:

kubectl apply -n NAMESPACE_NAME 
-f FILE_NAME

You can also use Terraform to manage your configurations.

To verify that TF Serving is emitting metrics on the expected endpoints, do the following:

Set up port forwarding by using the following command:

kubectl -n NAMESPACE_NAME 
port-forward POD_NAME 
8000

Access the endpoint localhost:8000/monitoring/prometheus/metrics by using the browser or the curl utility in another terminal session.

Define a PodMonitoring resource

For target discovery, the Managed Service for Prometheus Operator requires a PodMonitoring resource that corresponds to TF Serving in the same namespace.

You can use the following PodMonitoring configuration:

  # Copyright 2025 Google LLC 
 # 
 # Licensed under the Apache License, Version 2.0 (the "License"); 
 # you may not use this file except in compliance with the License. 
 # You may obtain a copy of the License at 
 # 
 #     https://www.apache.org/licenses/LICENSE-2.0 
 # 
 # Unless required by applicable law or agreed to in writing, software 
 # distributed under the License is distributed on an "AS IS" BASIS, 
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 
 # See the License for the specific language governing permissions and 
 # limitations under the License. 
 apiVersion 
 : 
  
 monitoring.googleapis.com/v1 
 kind 
 : 
  
 PodMonitoring 
 metadata 
 : 
  
 name 
 : 
  
 tfserve 
  
 labels 
 : 
  
 app.kubernetes.io/name 
 : 
  
 tfserve 
  
 app.kubernetes.io/part-of 
 : 
  
 google-cloud-managed-prometheus 
 spec 
 : 
  
 endpoints 
 : 
  
 - 
  
 port 
 : 
  
 8000 
  
 scheme 
 : 
  
 http 
  
 interval 
 : 
  
 30s 
  
 path 
 : 
  
 /monitoring/prometheus/metrics 
  
 selector 
 : 
  
 matchLabels 
 : 
  
 app 
 : 
  
 tfserve

To apply configuration changes from a local file, run the following command:

kubectl apply -n NAMESPACE_NAME 
-f FILE_NAME

You can also use Terraform to manage your configurations.

Verify the configuration

You can use Metrics Explorer to verify that you correctly configured TF Serving. It might take one or two minutes for Cloud Monitoring to ingest your metrics.

To verify the metrics are ingested, do the following:

In the Google Cloud console, go to the Metrics explorer page:

Go to Metrics explorer

If you use the search bar to find this page, then select the result whose subheading is Monitoring .
In the toolbar of the query-builder pane, select the button whose name is either MQL or PromQL .
Verify that PromQL is selected in the Language toggle. The language toggle is in the same toolbar that lets you format your query.

Enter and run the following query:

up{job="tfserve", cluster=" CLUSTER_NAME 
", namespace=" NAMESPACE_NAME 
"}

View dashboards

The Cloud Monitoring integration includes the TensorFlow Serving Prometheus Overview dashboard. Dashboards are automatically installed when you configure the integration. You can also view static previews of dashboards without installing the integration.

To view an installed dashboard, do the following:

In the Google Cloud console, go to the Dashboards page:

Go to Dashboards

If you use the search bar to find this page, then select the result whose subheading is Monitoring .
Select the Dashboard List tab.
Choose the Integrations category.
Click the name of the dashboard, for example, TensorFlow Serving Prometheus Overview .

To view a static preview of the dashboard, do the following:

In the Google Cloud console, go to the Integrations page:

Go to Integrations

If you use the search bar to find this page, then select the result whose subheading is Monitoring .
Click the Kubernetes Engine deployment-platform filter.
Locate the TensorFlow Serving integration and click View Details .
Select the Dashboards tab.

Troubleshooting

For information about troubleshooting metric ingestion problems, see Problems with collection from exporters in Troubleshooting ingestion-side problems .

TensorFlow Serving Stay organized with collections Save and categorize content based on your preferences.