A cluster-ready testlab, suitable for monitoring the behavior of p2p systems at scale. Built on nomad and consul.
🚧 This project is under active development! 🚧
Check out the ROADMAP to see what's coming.
- Dependencies
- Installation
- How it Works
- Usage
- Contribute
- Help Wanted
- License
You'll need a nomad cluster which, in turn, requires a consul deployment, in order to run testlab.
In development, the configuration files in
automation/packer/config
should be sufficient to run a single node
deployment. Furthermore, the packer configuration, with the help of the
Makefile
can build a simple VM image for either VMWare or VirtualBox,
featuring a testlab binary. Try the commands:
$ make vm-virtualbox
or
make vm-vmware
When deploying nomad manually, you must take care to deploy the nomad agent as root, since it requires on cgroups and/or docker to launch sandboxed tasks.
In production, a larger nomad deployment is advised. Hashicorp has recipes for
deploying nomad clusters on aws.
In the near future, testlab will include its own terraform recipes in the
automation
directory.
Testlab is a simple go binary, and can be installed into your GOPATH as such:
$ go get github.com/libp2p/testlab/testlab
Testlab is an automation layer over Hashicorp's Nomad, a cluster manager in the same style as Kubernetes. Testlab's primary goal is to make it simple to launch large clusters of peer-to-peer applications to better understand how they function at scale.
Testlab topologies are built around two main concepts: peer deployments and scenario runners. Generally, a peer deployment describes a set of instances of a peer-to-peer application and, optionally, how they are connected. A scenario runner is a special program launched in the cluster that can remotely control peer deployments to simulate activity within the network.
The goal output of a testlab topology is metrics data. While, in the future,
it would be nice to support correctness tests, the current aim is to allow for
large scale benchmarking and diagnosis of issues, as well as regression
testing. All peer deployments should be instrumented with prometheus-friendly
metrics, should they want to have data collected. This is described in greater
detail in the scenario runners section. Clusters
specifying a deployment of the prometheus
plugin will automatically have
metrics collected.
Testlab is a wrapper over nomad's golang API, making it easy to deploy pre-configured networks of p2p applications.
Most users of testlab need only concern themselves with two concepts, the deployment configuration, and scenarios. Users wishing to add testlab support for their own daemons will need to understand the node API, as well.
The testlab CLI depends on the presence of the standard environment variables to connect out to your Nomad and Consul clusters. If any are ommitted, the defaults, as defined by Hashicorp, will be applied. The defaults are typically usable in development.
Furthermore, users can optionally provide a path in the environment variable
TESTLAB_ROOT
to define where the testlab metadata will be stored. This
defaults to /tmp/testlab
. NOTE: In order to have multiple testlab
topologies in flight at the same time, one must define different TESTLAB_ROOT
s
for each topology. This requirement exists as a result of testlab associating a
single nomad deployment ID with each TESTLAB_ROOT
, though this can be extended
quite easily in the future.
The testlab CLI has two commands:
testlab start <json configuration>
Parses, evaluates for correctness, and attempts to deploy a topology as defined by the provided json configuration file. Once all of the peer-to-peer nodes a scenario depends on are deployed, the scenario will be deployed.testlab stop
Stops the current running topology, identified by itsTESTLAB_ROOT
.
The entrypoint for most projects using the testlab will be their deployment configuration, a JSON document declaring the desired network configuration. An example config can be found in the examples directory. It's broken into the following top level sections:
The name of the deployment. This will become a prefix to all tasks launched in the testlab.
Cluster-wide options to apply to the deployment.
{
// Datacenters is a list of nomad datacenters on which this test deployment
// should be scheduled. Nomad supports multiple datacenter deployments. By
// default this should be all datacenters.
"Datacenters": list of strings,
// Priority is an integer from [1, 100], the higher the more important. This
// allows nomad to determine which tasks should be scheduled when there is
// resource contention. If your nomad cluster has other tasks running on it,
// be sure to set this value accordingly. Otherwise, a default of 50 will be
// provided.
"Priority": int,
}
The deployments are where it gets interesting! Each deployment defines a class
of node to be scheduled on the cluster. Each deployment must define a
Name
, Plugin
, and Quantity
and may optionally define
Options
specific to the plugin and Dependencies
.
The name of this set of peers. This name will be used to reference these peers
in the Dependencies
.
Defines which node plugin to use. This defines how these nomad tasks will be configured. Must be one of the string identifiers listed in the node implementations section.
Defines how many of this type of peer should be launched in the cluster.
An optional object as defined by the specific node implementation.
A list of Name
s of deployments that must be scheduled before this one.
This feature exists for many reasons, such as allowing gateway nodes to go up
before generic peers that might want to bootstrap on them, or ensuring a
deployment of peers is launched before. The scenario that drives them is
scheduled. Cycles are not permitted.
Scenario runners are the beating heart of testlab's simulation capabilities.
It is their responsibility to drive the various deployments to create activity
within the network. While it's not entirely necessary to use the scenario
node
to deploy a scenario runner, it can be quite useful, especially in larger
clusters.
The scenario runner API is described by its node implementation and is, at present, a work in progress. Pull requests welcome!
Scenario runners can expect a few environment variables to be present, to aid them in connecting to the peers they wish to control. These variables are mostly tailored towards helping them interact with Consul, to discover information about the peers they've been assigned to.
DAEMON_CLIENTS
(int): The number of TCP/UDP ports this scenario runner has been allocated. These ports can be used for callbacks from daemons, such as how the libp2p daemon uses callbacks to receive incoming streams, etc. TODO: This should be become a more generic key, likeTESTLAB_PORTS
.SERVICE_TAG
(string): The tag that will be applied to the Consul services this runner is meant to control. For example, if a scenario is controlling libp2p daemons, which expose ap2pd
service for daemon control, it could query the consul cluster forp2pd
services with the$SERVICE_TAG
tag, yielding the daemon control port of every daemon under their purview.CONSUL_*
(various): Additionally, the standard set of consul environment variables will be present, so that the scenario may connect to the consul cluster.
As will be documented below in the node implementations
section, users can pass in any additional environment variables they wish to
their scenario runner via the Env
option in their configuration.
This set of environment variables is the extent of the scenario runner "API". It is up to the user how to use these. If working in golang, one can use the nascent golang scenario runner API, which provides convenience functions for accessing consul and creating libp2p daemon clients. TODO: Generalize this library to focus entirely on consul access, and split libp2p specific functionality into a separate sub-package.
Nodes describe how peer-to-peer applications should be launched within the cluster. In order to add testlab support for your peer-to-peer application, you must implement the following api
package node
import (
capi "github.com/hashicorp/consul/api"
napi "github.com/hashicorp/nomad/api"
utils "github.com/libp2p/testlab/utils"
)
type Node interface {
Task(utils.NodeOptions) (*napi.Task, error)
PostDeploy(*capi.Client, utils.NodeOptions) error
}
Given some utils.NodeOptions
, a wrapper over the map[string]interface{}
type
generated by JSON deserialization in go, a Node
must generate a
Nomad task or
return an error.
Furthermore, a Node
must implement a post-deployment hook (can be no-op), a
function that is called after deployments of this type have been successfully
scheduled in the cluster. This can be useful for connecting to the newly
launched peers and writing important metadata pertaining to them into Consul's
KV store. An example of this is the libp2p daemon, which uses it to associate
a peer's randomly generated ID with it's consul service ID.
At present, there are three node implementations:
p2pd
: the libp2p daemonscenario
: the generic scenario runnerprometheus
: prometheus metrics collection
A description of their behavior and configuration options follows.
The p2pd plugin adds support for the libp2p daemon. It will spawn libp2p peers, exposing the following services:
libp2p
: The libp2p host.p2pd
: The libp2p daemon control endpoint, exposed so scenario runners can manipulate the peer.metrics
: Prometheus scraping endpoint.
libp2p daemons can be configured with the following options:
PubsubRouter
string (optional): "gossipsub" or "floodsub", per users preference.Cid
string (optional): instead of looking for thep2pd
binary on the local filesystem, testlab can fetch a binary from IPFS by it's Cid.Fetch
string (optional): instead of looking for thep2pd
binary on the local filesystem, testlab can fetch a binary from an arbitrary (http/s) URL.Tags
list of strings (optional): Tags to apply to the service entries in Consul. These make it possible for scenarios to reference the specific subset of peers they're assigned to manipulate.Bootstrap
string (optional): The name of another deployment representing the network's "bootstrapper" (well known entrypoint) nodes. These will be automatically connected to when the daemon starts.
After the libp2p daemons are successfully scheduled on the cluster, testlab will
query each peer for its peer ID and store it in the Consul KV store under the
key "peerid/<multiaddr to libp2p service>"
e.g. peerid/ip4/127.0.0.1/tcp/6
.
The scenario plugin adds support for launching scenario runners in the testlab cluster. They must either be present on the clusters /usr/... path, or can be fetched from a URL like the libp2p daemon. Scenario runners will be provided environment variables as described above.
Scenario runners can be configured with the following options:
Clients
int (required): The number of TCP/UDP ports to allocate for this scenario. So-named because the libp2p daemon requires ports in order to receive information pushed from the daemon. TODO: Generalize this.Fetch
string (optional): instead of looking for thep2pd
binary on the local filesystem, testlab can fetch a binary from an arbitrary (http/s) URL.
None.
The prometheus plugin adds support for launching a
Prometheus metrics collector. Testlab automatically
configures prometheus to scrape Consul for all tasks exposing a metrics
service.
NOTE: As previously mentioned, all CONSUL_*
and NOMAD_*
environment
variables must be defined in the terminal that testlab
is executed from. If
they are not, they will not be passed along to the prometheus configuration.
This can result in prometheus failing to scrape Consul.
NOTE: Currently, a prometheus node still needs to be manually added to the topology configuration. This may become automatic in the future.
None.
None.
Feel free to join in. All welcome. Open an issue!
This repository falls under the IPFS Code of Conduct.
If you've got a peer-to-peer application you'd like to start testing and
benchmarking at scale, don't hesitate to submit a PR adding a Node
for it!
Please feel free to ask any questions in the issues or on #libp2p on freenode.
MIT / Apache 2 Dual License