Local troubleshooting

This tutorial shows how to troubleshoot a broken Knative serving service using Stackdriver tools for discovery and a local development workflow for investigation.

This step-by-step "case study" companion to the troubleshooting guide uses a sample project that results in runtime errors when deployed, which you troubleshoot to find and fix the problem.

Objectives

  • Write, build, and deploy a service to Knative serving
  • Use Cloud Logging to identify an error
  • Retrieve the container image from Container Registry for a root cause analysis
  • Fix the "production" service, then improve the service to mitigate future problems

Costs

In this document, you use the following billable components of Google Cloud:

To generate a cost estimate based on your projected usage, use the pricing calculator. New Google Cloud users might be eligible for a free trial.

Before you begin

Assembling the code

Build a new Knative serving greeter service step-by-step. As a reminder, this service creates a runtime error on purpose for the troubleshooting exercise.

  1. Create a new project:

    Node.js

    Create a Node.js project by defining the service package, initial dependencies, and some common operations.

    1. Create a hello-service directory:

      mkdir hello-service
cd hello-service

    2. Generate a package.json file:

      npm init --yes
npm install --save express@4

    3. Open the new package.json file in your editor and configure a start script to run node index.js. When you're done, the file will look like this:

      {
  "name": "hello-service",
  "version": "1.0.0",
  "description": "",
  "main": "index.js",
  "scripts": {
      "start": "node index.js",
      "test": "echo \"Error: no test specified\" && exit 1"
  },
  "keywords": [],
  "author": "",
  "license": "ISC",
  "dependencies": {
      "express": "^4.17.1"
  }
}

    If you continue to evolve this service beyond the immediate tutorial, consider filling in the description, author, and evaluate the license. For more details, read the package.json documentation.

    Python

    1. Create a new hello-service directory:

      mkdir hello-service
cd hello-service

    2. Create a requirements.txt file and copy your dependencies into it: 

      Flask==3.0.3
pytest==8.2.0; python_version > "3.0"
# pin pytest to 4.6.11 for Python2.
pytest==4.6.11; python_version < "3.0"
gunicorn==22.0.0
Werkzeug==3.0.3

    Go

    1. Create a hello-service directory:

      mkdir hello-service
cd hello-service

    2. Create a Go project by initializing a new go module:

      go mod init <var>my-domain</var>.com/hello-service

    You can update the specific name as you wish: you should update the name if the code is published to a web-reachable code repository.

    Java

    1. Create a maven project:

      mvn archetype:generate \
  -DgroupId=com.example \
  -DartifactId=hello-service \
  -DarchetypeArtifactId=maven-archetype-quickstart \
  -DinteractiveMode=false

    2. Copy the dependencies into your pom.xml dependency list (between the <dependencies> elements):

      <dependency>
  <groupId>com.sparkjava</groupId>
  <artifactId>spark-core</artifactId>
  <version>2.9.4</version>
</dependency>
<dependency>
  <groupId>org.slf4j</groupId>
  <artifactId>slf4j-api</artifactId>
  <version>2.0.12</version>
</dependency>
<dependency>
  <groupId>org.slf4j</groupId>
  <artifactId>slf4j-simple</artifactId>
  <version>2.0.12</version>
</dependency>

    3. Copy the build setting into your pom.xml (under the <dependencies> elements):

      <build>
  <plugins>
    <plugin>
      <groupId>com.google.cloud.tools</groupId>
      <artifactId>jib-maven-plugin</artifactId>
      <version>3.4.0</version>
      <configuration>
        <to>
          <image>gcr.io/PROJECT_ID/hello-service</image>
        </to>
      </configuration>
    </plugin>
  </plugins>
</build>

  2. Create an HTTP service to handle incoming requests:

    Node.js 

    const express = require('express');
const app = express();

app.get('/', (req, res) => {
  console.log('hello: received request.');

  const {NAME} = process.env;
  if (!NAME) {
    // Plain error logs do not appear in Stackdriver Error Reporting.
    console.error('Environment validation failed.');
    console.error(new Error('Missing required server parameter'));
    return res.status(500).send('Internal Server Error');
  }
  res.send(`Hello ${NAME}!`);
});
const port = parseInt(process.env.PORT) || 8080;
app.listen(port, () => {
  console.log(`hello: listening on port ${port}`);
});

    Python 

    import json
import os

from flask import Flask


app = Flask(__name__)


@app.route("/", methods=["GET"])
def index():
    """Example route for testing local troubleshooting.

    This route may raise an HTTP 5XX error due to missing environment variable.
    """
    print("hello: received request.")

    NAME = os.getenv("NAME")

    if not NAME:
        print("Environment validation failed.")
        raise Exception("Missing required service parameter.")

    return f"Hello {NAME}"


if __name__ == "__main__":
    PORT = int(os.getenv("PORT")) if os.getenv("PORT") else 8080

    # This is used when running locally. Gunicorn is used to run the
    # application on Cloud Run. See entrypoint in Dockerfile.
    app.run(host="127.0.0.1", port=PORT, debug=True)

    Go 

    
// Sample hello demonstrates a difficult to troubleshoot service.
package main

import (
	"fmt"
	"log"
	"net/http"
	"os"
)

func main() {
	log.Print("hello: service started")

	http.HandleFunc("/", helloHandler)


	port := os.Getenv("PORT")
	if port == "" {
		port = "8080"
		log.Printf("Defaulting to port %s", port)
	}

	log.Printf("Listening on port %s", port)
	log.Fatal(http.ListenAndServe(fmt.Sprintf(":%s", port), nil))
}

func helloHandler(w http.ResponseWriter, r *http.Request) {
	log.Print("hello: received request")

	name := os.Getenv("NAME")
	if name == "" {
		log.Printf("Missing required server parameter")
		// The panic stack trace appears in Cloud Error Reporting.
		panic("Missing required server parameter")
	}

	fmt.Fprintf(w, "Hello %s!\n", name)
}

    Java 

    import static spark.Spark.get;
import static spark.Spark.port;

import org.slf4j.Logger;
import org.slf4j.LoggerFactory;

public class App {

  private static final Logger logger = LoggerFactory.getLogger(App.class);

  public static void main(String[] args) {
    int port = Integer.parseInt(System.getenv().getOrDefault("PORT", "8080"));
    port(port);

    get(
        "/",
        (req, res) -> {
          logger.info("Hello: received request.");
          String name = System.getenv("NAME");
          if (name == null) {
            // Standard error logs do not appear in Stackdriver Error Reporting.
            System.err.println("Environment validation failed.");
            String msg = "Missing required server parameter";
            logger.error(msg, new Exception(msg));
            res.status(500);
            return "Internal Server Error";
          }
          res.status(200);
          return String.format("Hello %s!", name);
        });
  }
}

  3. Create a Dockerfile to define the container image used to deploy the service:

    Node.js 

    
# Use the official lightweight Node.js image.
# https://hub.docker.com/_/node
FROM node:20-slim
# Create and change to the app directory.
WORKDIR /usr/src/app

# Copy application dependency manifests to the container image.
# A wildcard is used to ensure copying both package.json AND package-lock.json (when available).
# Copying this first prevents re-running npm install on every code change.
COPY package*.json ./

# Install dependencies.
# if you need a deterministic and repeatable build create a
# package-lock.json file and use npm ci:
# RUN npm ci --omit=dev
# if you need to include development dependencies during development
# of your application, use:
# RUN npm install --dev

RUN npm install --omit=dev

# Copy local code to the container image.
COPY . ./

# Run the web service on container startup.
CMD [ "npm", "start" ]

    Python 

    
# Use the official Python image.
# https://hub.docker.com/_/python
FROM python:3.11

# Allow statements and log messages to immediately appear in the Cloud Run logs
ENV PYTHONUNBUFFERED True

# Copy application dependency manifests to the container image.
# Copying this separately prevents re-running pip install on every code change.
COPY requirements.txt ./

# Install production dependencies.
RUN pip install -r requirements.txt

# Copy local code to the container image.
ENV APP_HOME /app
WORKDIR $APP_HOME
COPY . ./

# Run the web service on container startup.
# Use gunicorn webserver with one worker process and 8 threads.
# For environments with multiple CPU cores, increase the number of workers
# to be equal to the cores available.
# Timeout is set to 0 to disable the timeouts of the workers to allow Cloud Run to handle instance scaling.
CMD exec gunicorn --bind :$PORT --workers 1 --threads 8 --timeout 0 main:app

    Go 

    
# Use the offical golang image to create a binary.
# This is based on Debian and sets the GOPATH to /go.
# https://hub.docker.com/_/golang
FROM golang:1.21-bookworm as builder

# Create and change to the app directory.
WORKDIR /app

# Retrieve application dependencies.
# This allows the container build to reuse cached dependencies.
# Expecting to copy go.mod and if present go.sum.
COPY go.* ./
RUN go mod download

# Copy local code to the container image.
COPY . ./

# Build the binary.
RUN go build -v -o server

# Use the official Debian slim image for a lean production container.
# https://hub.docker.com/_/debian
# https://docs.docker.com/develop/develop-images/multistage-build/#use-multi-stage-builds
FROM debian:bookworm-slim
RUN set -x && apt-get update && DEBIAN_FRONTEND=noninteractive apt-get install -y \
    ca-certificates && \
    rm -rf /var/lib/apt/lists/*

# Copy the binary to the production image from the builder stage.
COPY --from=builder /app/server /server

# Run the web service on container startup.
CMD ["/server"]

    Java

    This sample uses Jib to build Docker images using common Java tools. Jib optimizes container builds without the need for a Dockerfile or having Docker installed. Learn more about building Java containers with Jib.

    <plugin>
  <groupId>com.google.cloud.tools</groupId>
  <artifactId>jib-maven-plugin</artifactId>
  <version>3.4.0</version>
  <configuration>
    <to>
      <image>gcr.io/PROJECT_ID/hello-service</image>
    </to>
  </configuration>
</plugin>

Shipping the code

Shipping code consists of three steps: building a container image with Cloud Build, uploading the container image to Container Registry, and deploying the container image to Knative serving.

To ship your code:

  1. Build your container and publish on Container Registry:

    Node.js 

    gcloud builds submit --tag gcr.io/PROJECT_ID/hello-service

    Where PROJECT_ID is your Google Cloud project ID. You can check your current project ID with gcloud config get-value project.

    Upon success, you should see a SUCCESS message containing the ID, creation time, and image name. The image is stored in Container Registry and can be re-used if desired.

    Python 

    gcloud builds submit --tag gcr.io/PROJECT_ID/hello-service

    Where PROJECT_ID is your Google Cloud project ID. You can check your current project ID with gcloud config get-value project.

    Upon success, you should see a SUCCESS message containing the ID, creation time, and image name. The image is stored in Container Registry and can be re-used if desired.

    Go 

    gcloud builds submit --tag gcr.io/PROJECT_ID/hello-service

    Where PROJECT_ID is your Google Cloud project ID. You can check your current project ID with gcloud config get-value project.

    Upon success, you should see a SUCCESS message containing the ID, creation time, and image name. The image is stored in Container Registry and can be re-used if desired.

    Java 

    mvn compile jib:build -Dimage=gcr.io/PROJECT_ID/hello-service

    Where PROJECT_ID is your Google Cloud project ID. You can check your current project ID with gcloud config get-value project.

    Upon success, you should see a BUILD SUCCESS message. The image is stored in Container Registry and can be re-used if desired.

  2. Run the following command to deploy your app:

    gcloud run deploy hello-service --image gcr.io/PROJECT_ID/hello-service

    Replace PROJECT_ID with your Google Cloud project ID. hello-service is both the container image name and name of the Knative serving service. Notice that the container image is deployed to the service and cluster that you configured previously under Setting up gcloud

    Wait until the deployment is complete: this can take about half a minute. On success, the command line displays the service URL.

Trying it out

Try out the service to confirm you have successfully deployed it. Requests should fail with a HTTP 500 or 503 error (members of the class 5xx Server errors). The tutorial walks through troubleshooting this error response.

If your cluster is configured with a routable default domain, skip the steps above and instead copy the URL into your web browser.

If you don't use automatic TLS certificates and domain mapping you are not provided a navigable URL for your service.

Instead, use the provided URL and the IP address of the service's ingress gateway to create a curl command that can make requests to your service:

  1. To obtain the external IP for the Load Balancer, run the following command:

    kubectl get svc istio-ingressgateway -n ASM-INGRESS-NAMESPACE

    Replace ASM-INGRESS-NAMESPACE with the namespace where your Cloud Service Mesh ingress is located. Specify istio-system if you installed Cloud Service Mesh using its default configuration.

    The resulting output looks similar to the following:

    NAME                   TYPE           CLUSTER-IP     EXTERNAL-IP  PORT(S)
istio-ingressgateway   LoadBalancer   XX.XX.XXX.XX   pending      80:32380/TCP,443:32390/TCP,32400:32400/TCP

    where the EXTERNAL-IP value is your external IP address of the Load Balancer.

  2. Run a curl command using this GATEWAY_IP address in the URL.

     curl -G -H "Host: SERVICE-DOMAIN" https://EXTERNAL-IP/

    Replace SERVICE-DOMAIN with the default assigned domain of your service. You can obtain this by taking the default URL and removing the protocol http://.

  3. See the HTTP 500 or HTTP 503 error message.

Investigating the problem

Visualize that the HTTP 5xx error encountered above in Trying it out was encountered as a production runtime error. This tutorial walks through a formal process for handling it. Although production error resolution processes vary widely, this tutorial presents a particular sequence of steps to show the application of useful tools and techniques.

To investigate this problem you will work through these phases:

  • Collect more details on the reported error to support further investigation and set a mitigation strategy.
  • Relieve user impact by deciding to push forward in a fix or rollback to a known-healthy version.
  • Reproduce the error to confirm the correct details have been gathered and that the error is not a one-time glitch
  • Perform a root cause analysis on the bug to find the code, configuration, or process which created this error

At the start of the investigation you have a URL, timestamp, and the message "Internal Server Error".

Gathering further details

Gather more information about the problem to understand what happened and determine next steps.

Use available tools to collect more details:

  1. View logs for more details.

  2. Use Cloud Logging to review the sequence of operations leading to the problem, including error messages.

Rollback to a healthy version

If you have a revision that you know was working, you can rollback your service to use that revision. For example, you will not be able to perform a rollback on the new hello-service service that you deployed in this tutorial because it contains only a single revision.

To locate a revision and rollback your service:

  1. List all of the revisions of your service.

  2. Migrate all traffic to the healthy revision.

Reproducing the error

Using the details you obtained previously, confirm the problem consistently occurs under test conditions.

Send the same HTTP request by trying it out again, and see if the same error and details are reported. It may take some time for error details to show up.

Because the sample service in this tutorial is read-only and doesn't trigger any complicating side effects, reproducing errors in production is safe. However, for many real services, this won't be the case: you may need to reproduce errors in a test environment or limit this step to local investigation.

Reproducing the error establishes the context for further work. For example, if developers cannot reproduce the error further investigation may require additional instrumentation of the service.

Performing a root cause analysis

Root cause analysis is an important step in effective troubleshooting to ensure you fix the problem instead of a symptom.

Previously in this tutorial, you reproduced the problem on Knative serving which confirms the problem is active when the service is hosted on Knative serving. Now reproduce the problem locally to determine if the problem is isolated to the code or if it only emerges in production hosting.

  1. If you have not used Docker CLI locally with Container Registry, authenticate it with gcloud:

    gcloud auth configure-docker

    For alternative approaches see Container Registry authentication methods.

  2. If the most recently used container image name is not available, the service description has the information of the most recently deployed container image:

    gcloud run services describe hello-service

    Find the container image name inside the spec object. A more targeted command can directly retrieve it:

    gcloud run services describe hello-service \
   --format="value(spec.template.spec.containers.image)"

    This command reveals a container image name such as gcr.io/PROJECT_ID/hello-service.

  3. Pull the container image from the Container Registry to your environment, this step might take several minutes as it downloads the container image:

    docker pull gcr.io/PROJECT_ID/hello-service

    Later updates to the container image that reuse this name can be retrieved with the same command. If you skip this step, the docker run command below pulls a container image if one is not present on the local machine.

  4. Run locally to confirm the problem is not unique to Knative serving:

    PORT=8080 && docker run --rm -e PORT=$PORT -p 9000:$PORT \
   gcr.io/PROJECT_ID/hello-service

    Breaking down the elements of the command above,

    • The PORT environment variable is used by the service to determine the port to listen on inside the container.
    • The run command starts the container, defaulting to the entrypoint command defined in the Dockerfile or a parent container image.
    • The --rm flag deletes the container instance on exit.
    • The -e flag assigns a value to an environment variable. -e PORT=$PORT is propagating the PORT variable from the local system into the container with the same variable name.
    • The -p flag publishes the container as a service available on localhost at port 9000. Requests to localhost:9000 will be routed to the container on port 8080. This means output from the service about the port number in use will not match how the service is accessed.
    • The final argument gcr.io/PROJECT_ID/hello-service is a repository path pointing to the latest version of the container image. If not available locally, docker attempts to retrieve the image from a remote registry.

    In your browser, open http://localhost:9000. Check the terminal output for error messages that match those on Google Cloud Observability.

    If the problem is not reproducible locally, it may be unique to the Knative serving environment. Review the Knative serving troubleshooting guide for specific areas to investigate.

    In this case the error is reproduced locally.

Now that the error is doubly-confirmed as persistent and caused by the service code instead of the hosting platform, it's time to investigate the code more closely.

For purposes of this tutorial it is safe to assume the code inside the container and the code in the local system is identical.

Node.js

Find the source of the error message in the file index.js around the line number called out in the stack trace shown in the logs: 
const {NAME} = process.env;
if (!NAME) {
  // Plain error logs do not appear in Stackdriver Error Reporting.
  console.error('Environment validation failed.');
  console.error(new Error('Missing required server parameter'));
  return res.status(500).send('Internal Server Error');
}

Python

Find the source of the error message in the file main.py around the line number called out in the stack trace shown in the logs: 
NAME = os.getenv("NAME")

if not NAME:
    print("Environment validation failed.")
    raise Exception("Missing required service parameter.")

Go

Find the source of the error message in the file main.go around the line number called out in the stack trace shown in the logs:

name := os.Getenv("NAME")
if name == "" {
	log.Printf("Missing required server parameter")
	// The panic stack trace appears in Cloud Error Reporting.
	panic("Missing required server parameter")
}

Java

Find the source of the error message in the file App.java around the line number called out in the stack trace shown in the logs:

String name = System.getenv("NAME");
if (name == null) {
  // Standard error logs do not appear in Stackdriver Error Reporting.
  System.err.println("Environment validation failed.");
  String msg = "Missing required server parameter";
  logger.error(msg, new Exception(msg));
  res.status(500);
  return "Internal Server Error";
}

Examining this code, the following actions are taken when the NAME environment variable is not set:

  • An error is logged to Google Cloud Observability
  • An HTTP error response is sent

The problem is caused by a missing variable, but the root cause is more specific: the code change adding the hard dependency on an environment variable did not include related changes to deployment scripts and runtime requirements documentation.

Fixing the root cause

Now that we have collected the code and identified the potential root cause, we can take steps to fix it.

  • Check whether the service works locally with the NAME environment available in place:

    1. Run the container locally with the environment variable added:

      PORT=8080 && docker run --rm -e PORT=$PORT -p 9000:$PORT \
 -e NAME="Local World!" \
 gcr.io/PROJECT_ID/hello-service

    2. Navigate your browser to http://localhost:9000

    3. See "Hello Local World!" appear on the page

  • Modify the running Knative serving service environment to include this variable:

    1. Run the services update command with the --update-env-vars parameter to add an environment variable:

      gcloud run services update hello-service \
  --update-env-vars NAME=Override

    2. Wait a few seconds while Knative serving creates a new revision based on the previous revision with the new environment variable added.

  • Confirm the service is now fixed:

    1. Navigate your browser to the Knative serving service URL.
    2. See "Hello Override!" appear on the page.
    3. Verify that no unexpected messages or errors appear in Cloud Logging.

Improving future troubleshooting speed

In this sample production problem, the error was related to operational configuration. There are code changes that will minimize the impact of this problem in the future.

  • Improve the error log to include more specific details.
  • Instead of returning an error, have the service fall back to a safe default. If using a default represents a change to normal functionality, use a warning message for monitoring purposes.

Let's step through removing the NAME environment variable as a hard dependency.

  1. Remove the existing NAME-handling code:

    Node.js 

    const {NAME} = process.env;
if (!NAME) {
  // Plain error logs do not appear in Stackdriver Error Reporting.
  console.error('Environment validation failed.');
  console.error(new Error('Missing required server parameter'));
  return res.status(500).send('Internal Server Error');
}

    Python 

    NAME = os.getenv("NAME")

if not NAME:
    print("Environment validation failed.")
    raise Exception("Missing required service parameter.")

    Go 

    name := os.Getenv("NAME")
if name == "" {
	log.Printf("Missing required server parameter")
	// The panic stack trace appears in Cloud Error Reporting.
	panic("Missing required server parameter")
}

    Java 

    String name = System.getenv("NAME");
if (name == null) {
  // Standard error logs do not appear in Stackdriver Error Reporting.
  System.err.println("Environment validation failed.");
  String msg = "Missing required server parameter";
  logger.error(msg, new Exception(msg));
  res.status(500);
  return "Internal Server Error";
}

  2. Add new code that sets a fallback value:

    Node.js 

    const NAME = process.env.NAME || 'World';
if (!process.env.NAME) {
  console.log(
    JSON.stringify({
      severity: 'WARNING',
      message: `NAME not set, default to '${NAME}'`,
    })
  );
}

    Python 

    NAME = os.getenv("NAME")

if not NAME:
    NAME = "World"
    error_message = {
        "severity": "WARNING",
        "message": f"NAME not set, default to {NAME}",
    }
    print(json.dumps(error_message))

    Go 

    name := os.Getenv("NAME")
if name == "" {
	name = "World"
	log.Printf("warning: NAME not set, default to %s", name)
}

    Java 

    String name = System.getenv().getOrDefault("NAME", "World");
if (System.getenv("NAME") == null) {
  logger.warn(String.format("NAME not set, default to %s", name));
}

  3. Test locally by re-building and running the container through the affected configuration cases:

    Node.js 

    docker build --tag gcr.io/PROJECT_ID/hello-service .

    Python 

    docker build --tag gcr.io/PROJECT_ID/hello-service .

    Go 

    docker build --tag gcr.io/PROJECT_ID/hello-service .

    Java 

    mvn compile jib:build

    Confirm the NAME environment variable still works:

    PORT=8080 && docker run --rm -e $PORT -p 9000:$PORT \
 -e NAME="Robust World" \
 gcr.io/PROJECT_ID/hello-service

    Confirm the service works without the NAME variable:

    PORT=8080 && docker run --rm -e $PORT -p 9000:$PORT \
 gcr.io/PROJECT_ID/hello-service

    If the service does not return a result, confirm the removal of code in the first step did not remove extra lines, such as those used to write the response.

  4. Deploy this by revisiting the Deploy your code section.

    Each deployment to a service creates a new revision and automatically starts serving traffic when ready.

    To clear the environment variables set earlier:

    gcloud run services update hello-service --clear-env-vars

Add the new functionality for the default value to automated test coverage for the service.

Finding other issues in the logs

You may see other issues in the Log Viewer for this service. For example, an unsupported system call will appear in the logs as a "Container Sandbox Limitation".

For example, the Node.js services sometimes result in this log message:

Container Sandbox Limitation: Unsupported syscall statx(0xffffff9c,0x3e1ba8e86d88,0x0,0xfff,0x3e1ba8e86970,0x3e1ba8e86a90). Please, refer to https://gvisor.dev/c/linux/amd64/statx for more information.

In this case, the lack of support does not impact the hello-service sample service.

Clean up

You can delete the resources created for this tutorial to avoid incurring costs.

Deleting tutorial resources

  1. Delete the Knative serving service you deployed in this tutorial:

    gcloud run services delete SERVICE-NAME

    Where SERVICE-NAME is your chosen service name.

    You can also delete Knative serving services from the Google Cloud console:

    Go to Knative serving

  2. Remove the gcloud default configurations you added during the tutorial setup:

     gcloud config unset run/platform
 gcloud config unset run/cluster
 gcloud config unset run/cluster_location

  3. Remove the project configuration:

     gcloud config unset project

  4. Delete other Google Cloud resources created in this tutorial:

What's next