Google App Engine — Bring the Code, Google Runs It

Google App Engine was launched as a preview on the evening of 7 April 2008, at a Google developer event called Campfire One. It let developers upload a web application and have Google run it on Google's own infrastructure — scaling it, storing its data, serving its traffic — without the developer renting or managing a single server. The preview was open to the first 10,000 developers to sign up, it was free within fixed quotas, and its only language was Python. Java followed in April 2009, and App Engine did not leave preview to become a fully supported Google product until November 2011.

In today's terms it was a platform as a service (PaaS): the customer brings code, the provider handles everything beneath it.

What Google announced

Google's App Engine blog opened that night with a post by product manager Paul McDonald, saying the preview release had been launched at "tonight's Campfire One". He described App Engine as a developer tool for running web applications on Google's infrastructure. Google's developer blog posted the same news after the event.

The preview came with:

The package included persistent storage built on Google's Bigtable and GFS, automatic scaling, Google APIs for user sign-in and email, and an SDK for running the application locally before uploading it.

Why Python

Three weeks later, a Google Developer Podcast episode with the App Engine team recorded that Python was chosen as the first language. The guests included tech lead Kevin Gibbs and Guido van Rossum, Python's creator, who by then worked at Google. Van Rossum explained that the runtime was the full Python language but that some libraries were missing, mostly for security reasons — App Engine ran many customers' code side by side and had to fence each one in.

That fence was the defining trade-off. Applications also stored their data in a schema-less datastore rather than a relational database — on the podcast Gibbs walked through how that differed in practice. In return, Google handled scaling.

App Engine against S3 and EC2

App Engine arrived two years after Amazon's first infrastructure services, and offered almost the opposite deal.

S3 and EC2 gave developers raw parts and left assembly to them. App Engine took the parts away and gave them a finished platform with rules. Both models survived; the question of how much of the stack the customer should see has run through every cloud service since.

Java, then out of preview

On 7 April 2009, a year to the day after launch, Google announced a second language. The post, titled "Seriously this time, the new language on App Engine: Java™", said Java had been the first and most popular request since launch. Java support opened as an early look to the first 10,000 developers, alongside scheduled tasks (cron), access to data behind corporate firewalls, and bulk import into the datastore.

App Engine stayed a preview for three and a half years. On 7 November 2011, with release 1.6.0, the App Engine team wrote that it had graduated from preview and was now a fully supported Google product. The same post:

• introduced a 99.95% SLA for paid applications using the High Replication Datastore
• reported more than 100 billion hits a month, more than 300,000 active applications and more than 100,000 developers — Google's own figures
• moved billing from CPU use to instance hours, a change that many developers questioned in the comments

What came after

In June 2014, on the day Google announced Kubernetes, Eric Brewer's post also announced extensions that let App Engine developers build and deploy Docker images in Managed VMs — a way out of the original sandbox for applications that needed more than it allowed. The same question shaped AWS Lambda in 2014, which took the hands-off model further, down to a single function.

The company is covered on Google, and the sequence on the cloud computing timeline.