gae: A Google AppEngine SDK wrapper

designed for testing and extensibility


(but generally it should be pretty stableish).



LUCI Go GAE adaptor code is meant to be worked on from an Chromium infra.git checkout, which enforces packages versions and Go toolchain version. First get fetch via depot_tools.git then run:

fetch infra
cd infra/go
eval `./`
cd src/


Contributing uses the same flow as Chromium contributions.

Expand ▾ Collapse ▴



Package gae provides a fakable wrapped interface for the appengine SDK's APIs. This means that it's possible to mock all of the supported appengine APIs for testing (or potentially implement a different backend for them).


gae currently provides interfaces for:

- Datastore
- Memcache
- TaskQueue
- Info (e.g. Namespace, AppID, etc.)

Additional features include:

- true service interfaces (not package-level functions)
- methods don't need explicit context passed to them, increasing readability.
- service filters allow for composition of functionality. For example:
  - transparent memcaching of datastore access
  - transparent transaction buffering
  - statistics-gathering shims
  - deterministic and probabalistic API failure simulation
  - transparent retries
- truly parallel in-memory testing implementation. No more need for subprocesses :).
- Separate service and user-facing interfaces
  - Allows easier filter and service implementation while retaining the
    benefits of a user-friendly interface.

Package Organization

The gae library is organized into several subpackages:

- service/*   supported service definitions
- impl/*      implementations of the services
- filter/*    extra filter functionality for the services, agnostic to the
              underlying implementation.

Service Definitions

A service defintion lives under the `service` subfolder, and defines the user-facing interface for a service. Each service has a few common types and functions. Common types are:

service.Interface    - the main user-friendly service interface.

service.RawInterface - the internal service interface used by service
                       and filter implementations. Note that some services
                       like Info don't distinguish between the service
                       interface and the user interface. This interface is
                       typically a bit lower level than Interface and
                       lacks convenience methods.

service.Testable     - any additional methods that a 'testing'
                       implementation should provide. This can be accessed
                       via the Testable method on RawInterface. If the
                       current implementation is not testable, it will
                       return nil. This is only meant to be accessed when

service.RawFactory   - a function returning a RawInterface

service.RawFilter    - a function returning a new RawInterface based on
                       the previous filtered interface. Filters chain
                       together to allow behavioral service features
                       without needing to agument the underlying service
                       implementations directly.

And common functions are:

service.GetRaw        - Retrieve the current, filtered RawInterface
                        implementation from the context. This is less
                        frequently used, but can be useful if you want to
                        avoid some of the overhead of the user-friendly
                        Interface, which can do sometimes-unnecessary amounts
                        of reflection or allocation. The RawInterface and
                        Interface for a service are fully interchangable and
                        usage of them can be freely mixed in an application.

service.AddRawFilters - adds one or more RawFilters to the context.

service.SetRawFactory - adds a RawFactory to the context

service.SetRaw        - adds an implementation of RawInterface to the context
                        (shorthand for SetRawFactory, useful for testing)


The impl subdirectory contains a couple different service implementations, depending on your needs.

'prod' is the production (e.g. real appengine-backed) implementation. It calls through to the original appengine SDK.

'memory' is a truly parallel in-memory testing implementation. It should be functionally the same as the production appengine services, implementing many of the real-world quirks of the actual services. It also implements the services' Testable interface, for those services which define those interfaces.

'dummy' provides a bunch of implementations of the various RawInterfaces. These implementations just panic with an appropriate message, depending on which API method was called. They're useful to embed in filter or service implementations as stubs while you're implementing the filter.


Each service also supports "filters". Filters are proxy objects which have the same interface as the service they're filtering, and pass data through to the previous filter in the stack. Conceptually, a filtered version of, for example, the Datastore, could look like:

User code
<count filter (counts how many times each API is called by the user)>
<dscache filter (attempts to use memcache as a cache for datastore)>
<count filter (counts how many times each API is actually hit)>
memory datastore.RawInterface implementation

Filters may or may not have state, it's up to the filter itself. In the case of the count filter, it returns its state from the Filter<Service> method, and the state can be observed to see how many times each API was invoked. Since filters stack, we can compare counts from rawCount versus userCount to see how many calls to the actual real datastore went through, vs. how many went to memcache, for example.

Note that Filters apply only to the service.RawInterface. All implementations of service.Interface boil down to calls to service.RawInterface methods, but it's possible that bad calls to the service.Interface methods could return an error before ever reaching the filters or service implementation.



package main

import (


func main() {
	// This is suitable for use in tests.
	c := memory.Use(context.Background())
	w := httptest.NewRecorder()
	_, err := http.NewRequest("GET", "/doesntmatter", nil)
	if err != nil {

	innerHandler(c, w)

// This is what you would use in production.
func handler(w http.ResponseWriter, r *http.Request) {
	c := context.Background()
	c = prod.Use(c, r)
	// add production filters, etc. here
	innerHandler(c, w)

type CoolStruct struct {
	ID string `gae:"$id"`

	Value string

func innerHandler(c context.Context, w http.ResponseWriter) {
	obj := &CoolStruct{ID: "struct-id", Value: "hello"}
	if err := datastore.Put(c, obj); err != nil {
		http.Error(w, err.Error(), http.StatusInternalServerError)
	fmt.Fprintf(w, "I wrote: %s", datastore.KeyForObj(c, obj))
I wrote: dev~app::/CoolStruct,"struct-id"

Source Files


Path Synopsis
filter/count Package count contains 'counter' filters for all the gae services.
filter/dscache Package dscache provides a transparent cache for RawDatastore which is backed by Memcache.
filter/featureBreaker Package featureBreaker contains filters for dynamically disabling/breaking API features at test-time.
filter/featureBreaker/flaky Package flaky can be used to emulate flaky GAE errors that happen randomly with some probability.
filter/readonly Package readonly implements a filter that enforces read-only accesses to services.
filter/txnBuf Package txnBuf contains a transaction buffer filter for the datastore service.
filter/txndefer Package txndefer implements a filter that calls best-effort callbacks on successful transaction commits.
impl/cloud Package cloud provides an implementation of services based on regular Cloud APIs and works from anywhere (not necessarily from AppEngine).
impl/dummy Package dummy provides panicking dummy implementations of all service Interfaces.
impl/memory Package memory provides an implementation of infra/gae/libs/wrapper which backs to local memory ONLY.
impl/prod Package prod provides an implementation of services/* which backs to AppEngine Classic (Go version <=1.11).
impl/prod/constraints Package constraints contains production datastore constraints.
service/blobstore Package blobstore is a PLACEHOLDER for the blobstore implementation.
service/datastore/dumper Package dumper implements a very VERY dumb datastore-dumping debugging aid.
service/datastore/internal/protos/datastore Package datastore is a generated protocol buffer package.
service/datastore/meta Package meta contains some methods for interacting with GAE's metadata APIs.
service/info/support Package support provides Info-related support functionality.
service/logging Package logging is simply documentation :) In order to use logging, please import and use the "" package.
service/urlfetch Package urlfetch provides a way for an application to get http.RoundTripper that can make outbound HTTP requests.