bigchaindb/proposals/integration-infrastructure.md

Integration testing simulation

Problem Description

The integration testing proposal introduced new set of requirements which would require that the infrastructure be controlled at the code level in order to write the tests.

Proposed Change

The following components shall be required

Network driver ND

The network driver is module which would implement the basic infrastructure management operations which would then be used in test cases to simulate different scenarios.

The Node class shall have the following methods:

• url : URL to access the BigchainDB node.
• start(): Start a node. This api will ensure that all components required for a fully functional node are up and running.
• stop(): Stop a given node.
• reset(): Reset the state of an existing node i.e. data stored in Tendermint and MongoDB will be purged. NOTE: It is only realistic to reset the nodes after all the tests have been executed rather than resetting the network after each test.
• start_tendermint(): Run Tendermint process on a given node.
• stop_tendermint(): Stop Tendermint process on a given node.
• configure_tendermint(): Alter Tendermint configuration on a given node.
• reset_tendermint(): Reset Tendermint i.e. tendermint unsafe_reset_all.
• start_db(): Start MongoDB daemon on a given node.
• stop_db(): Stop MongoDB daemon on a given node.
• reset_db(): Flush Bigchain collection from MongoDB.
• configure_bigchaindb(): Update BigchainDB configuration.
• configure_clock(): The method should allow to change the clock on a given node. This could facilitate simulation of nodes located in different time zones.
• configure_QoS(): Configure different scenarios like network latency, packet loss etc. The underlying implementation could be facilitated via tc.

The network driver implementation could be achieved using the following libraries

• Kubernetes Client: https://github.com/kubernetes-incubator/client-python
• Docker Client: https://github.com/docker/docker-py

Usage example

def test_valid_transaction_is_synced_nodes(network):
    alice, bob = crypto.generate_key_pair(), crypto.generate_key_pair()
    tx_alice_to_bob = Transaction.create([alice.public_key], [([bob.public_key], 1)])\
                                 .sign([alice.private_key])
    tx = tx.sign([user_priv])
    network.start(4)
    network.nodes[0].stop()
    requests.post(network.node[1].url + 'transactions', data=json.dumps(tx.to_dict()))
    network.node[0].start()

    time.sleep(1.5)

    for node in network.nodes:
        assert requests.get(node.url + 'transactions/' + tx.id) == tx

    network.stop()

A detailed list of testing scenarios can be found here.

Kubernetes Client resources

• Create Pod
• Delete Pod
• Create Node
• Delete Node
• Exec Command post, Exec Command get

Setup integration test environment

Kubernetes can be run locally using minikube but the default parameters used when running Kubernetes via minikube start a single-node Kubernetes cluster using a virtual machine which results in major resource consumption. This could be avoided by using command line flag, minikube start --vm-driver=none. For more details refer here.

Data model impact

N/A

API impact

N/A

Security impact

N/A

Performance impact

N/A

End user impact

N/A

Deployment impact

N/A

Documentation impact

N/A

Testing impact

The network driver would facilitate the design and implementation of components essential for integration testing. But this introduces Kubernetes as a dependency for tests which should be installed and configured before running tests on Travis CI (refer here).

Implementation

Assignee(s)

Primary assignee(s): @kansi

Targeted Release

BigchainDB 2.0

Dependencies

N/A

Reference(s)

N/A