Skip to content

Latest commit

 

History

History
131 lines (98 loc) · 5.01 KB

google-cloud-container-engine-first-impressions.md

File metadata and controls

131 lines (98 loc) · 5.01 KB
Raw

Google Cloud Container Engine - First impressions

Run Docker containers on Google Cloud Platform, powered by Kubernetes.

What they promise:

  • Easy container management and setup
  • Run docker containers in cloud
  • JSON to configure CPU, memory etc.
  • Logging
  • Reserved IP ranges
  • Free for up to 5 nodes
  • Open source tools

Sounds good, let's get started by creating a new project, but first .. what's Kubernetes?

Manage a cluster of Linux containers as a single system to accelerate Dev and simplify Ops.

Ok, an open source thingy.. woah 24 527 commits by 650 contributors. Must be a serious thing.

Continuing with the setup, account billing and Container Engine API need to be enabled before using it. Let's create a new project called busse-api. That project is currently running on Heroku, so I can easily compare them. Google let's you to choose the server node location from us-central, us-east1 or europe-west. I'll choose the europe-west one.

After this the mangement UI says "it takes a few minutes to get the Container Engine ready". In the meanwhile, let's install the needed tools. They provide a basic "pipe content from our site to bash" -installation, which I will happily run. It added a few things to my .zshrc, after a confirmation.

Running gcloud init pops up a browser tab, where I can give Google Cloud Platform a bunch of rights when logging in with my Google credentials. It redirects that back to localhost, so apparently gcloud command setup a temporary local http server for the trick, neat. It then asks you for a few default settings.

The kubectl can be installed with gcloud components install kubectl. Now the management UI has done its background work, let's create a container cluster. Let's name it busse-api too. I choose two of the laziest micro machines to europe-west-b1 zone. Error, you need to have 3 machines in the cluster with the chosen machine type. No possibility to edit - delete and back. It was really fast, so no problem. Second try. It takes a while to create the cluster.

After it's done, we can configure it to command line:

gcloud config set container/cluster busse-api
# It matters what you set in the compute zone in the previous gloud init command
gcloud container clusters get-credentials busse-api

Ok now the cluster of computers should be running in Google Cloud. Now I'm at the same point what Heroku provides me, almost. Next, deploying the Node project to the cluster.

cd busse-api
export PROJECT_ID=your-project-id-here

Create Dockerfile. They are advising quite old version of node, let's define our own based on nodejs.org guide.

FROM node:argon

# Create app directory
RUN mkdir -p /usr/src/app
WORKDIR /usr/src/app

# Install app dependencies
COPY package.json /usr/src/app/
RUN npm install

# Bundle app source
COPY . /usr/src/app

PORT=8080
# Still using Babel 5 in this project..
CMD [ "./node_modules/.bin/babel-node", "./src/index.js" ]

Push that to the Google Container Repository:

docker build -t gcr.io/${PROJECT_ID}/busse-api:v1 .
gcloud docker push gcr.io/${PROJECT_ID}/busse-api:v1

While running the second command, it returned 500 Internal Server Error once

Create a pod:

kubectl run busse-api --image=gcr.io/${PROJECT_ID}/busse-api:v1 --port=8080

At this point I got somd build errors first time because of a problem in my Dockerfile. I hadn't told it to run npm install. Checking the error logs was quite a challenge as I wasn't sure how to do it. Fortunately Google has a debugging pods section in their docs. After 5 minutes of trying different commands, I got some logs showing. In general, this shows what is happening in your pod: kubectl describe pods. In the end I got the original error with this command: kubectl logs --tail=100 --previous busse-api-1c2dl busse-api. Where busse-api-1c2dl is the Name when running describe pods.

After the Dockerfile fixes and repushing the image to Google container repo, I got my app running, almost! Describe pods says "Back-off restarting failed docker container". Ok, so let's check the logs from the container:

Error: MTA_API_KEY variable is not set!
    at Object.<anonymous> (/usr/src/app/src/init-env-variables.js:11:11)
    at Module._compile (module.js:410:26)
    ...

It's at least running my code already but now I need to specify the environment variables. This leads us to Kubernetes Secrets documentation.

THIS IS WHERE I LOST MY INTEREST..

The command-line tool (kubectl) required for Google Container Engine is supported on Linux and Mac OS X. Windows users can start a Google Compute Engine instance to access a Linux environment from which to run the tool.