Zmey is a micro-framework for testing distributed algorithms.
Zmey hides the complexity of setting up and running a distributed system; and managing inter-process and client-process communication. You're only required to implement a simple interface, telling how the process should react to incoming messages and client calls; and the injector, supplying the initial set of messages. Zmey will then run the implemenation of the algorithms at a given scale and collect the returned messages.
The best place to use Zmey is unit tests -- it is fast and quite deterministic:
func TestProcess(t *testing.T) {
// Create Zmey instance
z := zmey.NewZmey(&zmey.Config{
Debug: true,
})
// Add 10 processes with ids from 0 to 9. Each process is created with
// NewProcess(pid int) function.
const scale = 10
for pid := 0; pid < scale; pid++ {
z.SetProcess(pid, NewProcess(pid))
}
// Prepare the slices of initial calls, one slice per process
initialCalls := make([][]interface{}, 10)
// ...
// Fill initialCalls
// ...
// Create inject function. An inject function receives process id and
// zmey.Client c, which is used to perform the calls from initialCalls
injectF := func(pid int, c zmey.Client) {
for k := 0; k < len(initialCalls[pid]); k++ {
c.Call(initialCalls[pid][k])
}
}
// Set inject function to be lazily executed
z.Inject(injectF)
// (optionally) Isolate process 3 from the rest of the network, and simulate its failure to other processes by creating and setting simple filter function.
filterF := func(from, to int) bool {
if from == 3 || to == 3 {
return false
}
return true
}
z.Filter(filterF)
ctx, _ := context.WithTimeout(context.Background(), 1*time.Minute)
// Run Zmey round. Processes' responses are captured and
// collected in actualOutput slices, one slice per corresponding process
actualOutput, _, _ := z.Round(ctx)
// You may need to sort actualOutput to make it more deterministic
expectedOutput := make([][]interface{}, 10)
// ...
// Fill expectedOutput
// ...
// Finally, compare expected and actual responses
assert.Equal(t, expectedOutput, actualOutput)
}The process being tested is created by NewProcess function. It should return an object, implementing zmey.Process interface:
type Process interface {
Init(
sendF func(to int, payload interface{}),
returnF func(payload interface{}),
traceF func(payload interface{}),
errorF func(error),
)
ReceiveNet(from int, payload interface{})
ReceiveCall(payload interface{})
Tick(uint)
}The framework notifies the process about incoming messages and client calls by calling ReceiveNet and ReceiveCall methods. Time passing is conveyed through Tick events. The process talks back to the framework through the set of callbacks, received via Init method.
sendF is used to communicate to other processes over the network, returnF -- to return the data back to the client. traceF and errorF are mainly used for logging. The returned data would be then collected and returned by the Round method.
For more details check out the forwarder example.
Zmey is in its alpha state. Current version is good for launching algorithms, and doing some failure simulation. It is capable of creating systems with different types of processes (client/sever, corrent/Byzantine server, etc), and doing some reconfiguration (adding, removing and replacing the processes). Next releases will primarily focus on stability and performance optimizations.
Also, the framework may include advanced network simulation, e.g. selective message drop and reordering.