Overview | Hive Commands | Simulators | Clients
Hive is a system for running integration tests against Ethereum clients.
In hive, integration tests are called 'simulations'. A simulation is controlled by a program (the 'simulator') written in any language. The simulator launches clients and contains test logic. It reports test results back to hive, where they are aggregated for display in a web browser.
What makes hive different from other, generic CI infrastructure is the tight integration of Ethereum clients and their features. Simulator programs usually don't need to care about the differences between client implementations because hive provides a common interface to launch and configure them all. At this time, clients can be configured for any Ethereum 1 network definition, i.e. genesis block and hard fork activation block numbers. Simulations can also instruct clients to load a pre-defined test chain and enable block mining. You can find more information about client configuration in the client documentation.
Ethereum Foundation operates a public instance of Hive to check for consensus compatibility, peer-to-peer networking spec compliance, and user API support for most Ethereum client implementations. You can find the latest test results at https://hivetests.ethdevops.io/.
This is an overview of some of the simulators which are currently implemented and running continuously on the production hive instance:
-
devp2p: This simulator runs 'eth', 'snap' and 'discv4' peer-to-peer protocol tests. The test suites themselves are maintained in the go-ethereum repository. In their hive adaptation, the simulator launches the client with a known test chain, obtains its peer-to-peer endpoint (theenode://URL) and sends protocol messages to it. The client's responses are analyzed by the test suite to ensure that they conform to the respective protocol specification. -
ethereum/sync: This simulator attempts to synchronize the blockchain among all clients. For each enabled client implementation, it creates one instance of the client as the 'source'. The 'source' client is initialized with a known test chain. The simulator then launches a 'sink' instance of every known client against the source and checks whether the sink can synchronize the chain from the source client. -
ethereum/consensus: This simulator runs the Ethereum 1 consensus tests against all clients. While client implementers are generally expected to run these tests themselves, they might not always run the latest tests, and may skip some of them if they take too long. Running these tests in a hive simulation ensures that none are skipped. -
ethereum/rpc: The RPC simulator configures a client for clique PoA mining and runs various tests against the web3 JSON-RPC interface. These tests ensure that the client is able to receive transactions via RPC, incorporate them into its chain, and report transaction results via the standard APIs. -
ethereum/graphql: This simulator initializes a client with a known test chain and enables the GraphQL API endpoint. It then performs certain queries and compares their output to known good outputs.
This section explains how a single simulation run works.
For a single run, the user provides the name of the simulator to run, and a set of client names to run against. For example:
./hive --sim ethereum/sync --client go-ethereum,besu,openethereum
Hive first builds simulator and client images using docker. It expects a Dockerfile in the
./simulators/ethereum/sync directory as well as a Dockerfile for each client (in
./clients/*/Dockerfile).
While the simulator build must always work without error, it's OK for some client builds to fail as long as one of them succeeds. This is because client code pulled from the respective upstream repositories may occasionally fail to build.
Once all images are built, the simulator program is launched in a docker container. The
HIVE_SIMULATOR environment variable contains the HTTP server URL of the hive controller.
The hive simulation API can be accessed at this URL. The simulator launches clients and
reports test results through the API.
When the simulator requests a client instance, the hive controller launches a new docker container using the built client image. The client container entry point receives configuration through environment variables and files provided by the simulator. Depending on this configuration data, the client entry point configures the client's genesis state and imports the test chain (if provided). The client is now expected to launch its network endpoints for RPC and p2p communication.
When the client has finished starting, the simulator program communicates with it on the RPC and p2p endpoints. More than one client may be launched, and the clients can also communicate with each other.
During the simulation run, information about 'test suites' and their test cases must be
provided by the simulator via the simulation API. The hive controller collects this
information in a JSON file. It also collects client logs as well as the output of the
simulator program. All files are written to the results directory (./workspace/logs).
When the simulator program exits, the simulator container and all client containers are
stopped and removed. The hive command then exits as well.
The results of simulation runs are stored in the 'result directory'. For every test suite executed by a simulator, a JSON file like the following is created:
{
"id": 0,
"name": "sync",
"description": "This test suite verifies that...",
"clientVersions": {
"besu": "",
"go-ethereum": ""
},
"simLog": "1612356621-simulator-a9a2e71a6aabe509bbde35c79e7f0ed9c259a642c19ba0da6167fa9efd0ea5a1.log"
"testCases": {
"1": {
"name": "besu as sync source",
"description": "This loads the test chain...",
"start": "2021-02-03T12:50:21.77396767Z",
"end": "2021-02-03T12:51:56.080650164Z",
"summaryResult": {
"pass": true,
"details": ""
},
"clientInfo": {
"893a6ea2": {
"ip": "172.17.0.4",
"name": "besu",
"instantiatedAt": "2021-02-03T12:51:04.371913809Z",
"logFile": "besu/client-893a6ea2.log"
}
}
}
}
}
The result directory also contains log files of simulator and client output.