Google App Engine Recipe
Google App Engine (GAE) provides a range of hosting options from pure PaaS (App Engine Classic) through Managed VMs to fully self-managed or container-driven Compute Engine instances. Echo works great with all of these but requires a few changes to the usual examples to run on the AppEngine Classic and Managed VM options. With a small amount of effort though it’s possible to produce a codebase that will run on these and also non-managed platforms automatically.
We’ll walk through the changes needed to support each option.
Wait? What? I thought this was about AppEngine! Bear with me - the easiest way to show the changes required is to start with a setup for standalone and work from there plus there’s no reason we wouldn’t want to retain the ability to run our app anywhere, right?
We take advantage of the go build constraints or tags to change how we create and run the Echo server for each platform while keeping the rest of the application (e.g. handler wireup) the same across all of them.
First, we have the normal setup based on the examples but we split it into two files -
be common to all variations and holds the Echo instance variable. We initialise it from a function
and because it is a
var this will happen before any
init() functions run - a feature that we’ll
use to connect our handlers later.
A separate source file contains the function to create the Echo instance and add the static
file handlers and middleware. Note the build tag on the first line which says to use this when not
bulding with appengine or appenginevm tags (which thoese platforms automatically add for us). We also
main() function to start serving our app as normal. This should all be very familiar.
The handler-wireup that would normally also be a part of this Echo setup moves to separate files which
take advantage of the ability to have multiple
init() functions which run after the
e Echo var is
initialised but before the
main() function is executed. These allow additional handlers to attach
themselves to the instance - I’ve found the
Group feature naturally fits into this pattern with a file
per REST endpoint, often with a higher-level
api group created that they attach to instead of the root
Echo instance directly (so things like CORS middleware can be added at this higher common-level).
If we run our app it should execute as it did before when everything was in one file although we have at least gained the ability to organize our handlers a little more cleanly.
AppEngine Classic and Managed VMs
So far we’ve seen how to split apart the Echo creation and setup but still have the same app that still only runs standalone. Now we’ll see hwo those changes allow us to add support for AppEngine hosting.
Refer to the AppEngine site for full configuration and deployment information.
app.yaml configuration file
Both of these are Platform as as Service options running on either sandboxed micro-containers
or managed Compute Engine instances. Both require an
app.yaml file to describe the app to
the service. While the app could still serve all it’s static files itself, one of the benefits
of the platform is having Google’s infrastructure handle that for us so it can be offloaded and
the app only has to deal with dynamic requests. The platform also handles logging and http gzip
compression so these can be removed from the codebase as well.
The yaml file also contains other options to control instance size and auto-scaling so for true
deployment freedom you would likely have separate
app-vm.yaml files and
this can help when making the transition from AppEngine Classic to Managed VMs.
We’ll now use the build constraints again like we did when creating
app-standalone.go instance but this time with the opposite tags to use this file if the build has
the appengine or appenginevm tags (added automatically when deploying to these platforms).
This allows us to replace the
createMux() function to create our Echo server without any of the
static file handling and logging + gzip middleware which is no longer required. Also worth nothing is
that GAE classic provides a wrapper to handle serving the app so instead of a
main() function where
we run the server, we instead wire up the router to the default
Managed VMs are slightly different. They are expected to respond to requests on port 8080 as well
as special health-check requests used by the service to detect if an instance is still running in
order to provide automated failover and instance replacement. The
package provides this for us so we have a slightly different version for Managed VMs:
So now we have three different configurations. We can build and run our app as normal so it can
be executed locally, on a full Compute Engine instance or any other traditional hosting provider
(including EC2, Docker etc…). This build will ignore the code in appengine and appenginevm tagged
files and the
app.yaml file is meaningless to anything other than the AppEngine platform.
Or Managed VMs:
gcloud config set project [your project id] gcloud preview app run .
And of course we can deploy our app to both of these platforms for easy and inexpensive auto-scaling joy.
Depending on what your app actually does it’s possible you may need to make other changes to allow switching between AppEngine provided service such as Datastore and alternative storage implementations such as MongoDB. A combination of go interfaces and build constraints can make this fairly straightforward but is outside the scope of this recipe.