name_manager

README

name_manager: Utility to manage shared test resources with a global lock 🔒

Tests should ideally set up and tear down all the resources they use. However, sometimes these resources are too expensive and they must be shared. For instance, instead of using a brand new database for each test, one can reuse the same database and truncate the tables between the tests. If one wants to run tests concurrently, two by two, one can create two databases and reuse them in the same way. name_manager can be used to share such test resources between processes and to avoid race conditions where two concurrent tests end up using the same shared resource.

name_manager has an obvious application for the concurrent allocation of shared test resources, as explained above, but its aim is more generic: it is a distributed system, with a global lock, to retrieve unique names while ensuring that previously acquired names, now released, are reused as much as possible.

name_manager can be either consumed as a Go package through a standalone Command-Line Interface (CLI).

Example use

By default, the name_manager uses the local file system as an inter-process lock. There is also a MongoDB backend, which can be used, e.g., with Azure CosmosDB.

As an example use, let's pretend we have a bunch of end-to-end tests on a webapp. These tests run on a full app that is deployed with docker-compose. Deploying with docker-compose is expensive, so we start by using a single docker-compose project and executing the tests serially:

# Creates the shared resource
docker-compose -p stack -f docker-compose.yml up -d

DOCKER_COMPOSE_PROJECT_NAME=stack test_1
DOCKER_COMPOSE_PROJECT_NAME=stack test_2
DOCKER_COMPOSE_PROJECT_NAME=stack test_3
DOCKER_COMPOSE_PROJECT_NAME=stack test_4

# Destroys the shared resource
docker-compose -f docker-compose.yml down

Note that, because the tests clean up the shared resource before use, docker-compose down is not necessary between invocations of the shell script above and will be omitted from now on.

If the tests all take the same time to complete, and we have enough system resources, we can more or less halve the execution time, outside of the setup/teardown steps, by providing two stacks and executing the tests two by two:

# Ensures the shared resources are up
docker-compose -p stack1 -f docker-compose.yml up -d &
docker-compose -p stack2 -f docker-compose.yml up -d &
wait

(
  DOCKER_COMPOSE_PROJECT_NAME=stack1 test_1 &&
  DOCKER_COMPOSE_PROJECT_NAME=stack1 test_2
) &
(
  DOCKER_COMPOSE_PROJECT_NAME=stack2 test_3 &&
  DOCKER_COMPOSE_PROJECT_NAME=stack2 test_4
) &
wait

name_manager enables a generalization of this reasoning. See the ./examples directory for examples.

Development

name_manager is compiled with Go 1.13.

The integration tests for the mongo backend require a MongoDB server to be running locally:

docker run --name some-mongo -p 27017:27017 mongo:bionic

export MONGODB_URI="mongodb://127.0.0.1:27017"
go test -v ./pkg/mongo_backend/...

The integration tests for the firestore backend require a Firestore emulator to be running locally:

docker run --name some-firestore -p 8080:8080 -e PORT=8080 ridedott/firestore-emulator:1.10.4

export FIRESTORE_EMULATOR_HOST="127.0.0.1:8080"
go test -v ./pkg/firestore_backend/...

License

name_manager is licensed under The MIT License.