6.0 KiB
layout, page_title, sidebar_current, description
| layout | page_title | sidebar_current | description |
|---|---|---|---|
| docs | Creating a Nomad Cluster | docs-cluster-bootstrap | Learn how to bootstrap a Nomad cluster. |
Creating a cluster
Nomad clusters in production comprises of a few Nomad servers (an odd number, preferrably 3 or 5, but never an even number to prevent split-brain), clients and optionally Consul servers and clients. Before we start discussing the specifics around bootstrapping clusters we should discuss the network topology. Nomad models infrastructure as Regions and Datacenters. Nomad Regions may contain multiple datacenters. Nomad Servers are assigned to the same region and hence a region is a single scheduling domain in Nomad. Each cluster of Nomad Servers support only one region, however multiple Regions can be stitched together to allow a globally coherent view of an organization's resources.
Nomad Servers
Nomad servers are expected to have network latencies no more than 10 milliseconds between them. Nomad servers could be spread across multiple datacenters, if they have low latency connections between them, to achieve high availability. For example, on AWS every region comprises of multiple zones which have very low latency links between them, so every zone can be modeled as a Nomad Data Center and every Zone can have a single Nomad server which could be connected to form a quorum and form a Region. Nomad servers uses raft for replicating state between them and raft being highly consistent needs a quorum of servers to function, therefore we recommend running an odd number of Nomad servers in a region. Usually running 3-5 servers in a region is recommended. The cluster can withstand a failure of one server in a cluster of three servers and two failures in a cluster of five servers. Adding more servers to the quorum adds more time to replicate state and hence throughput decreases so we don't recommend having more than seven servers in a region.
During the bootstrapping phase Nomad servers need to know the addresses of other
servers, which can be achieved by using the -reconnect-join cli command or by
pointing Nomad servers to a Consul Service.
Nomad Clients
Nomad clients are organized in DCs and they have to be seeded by the list of servers that they will have to connect.
Operators can either place the addresses of the Nomad servers in the client configuration or point Nomad client to the Nomad server service in Consul. Once a client establishes connection with a Nomad servers, if new servers are added to the cluster the addresses are propagated down to the clients along with heartbeat.
Consul Cluster
Bootstrapping a Nomad cluster becomes significantly easier if operators use Consul and registers Nomad servers with Consul. Network topology of a Consul cluster is slightly different than Nomad. Consul models infrastructures as Data Centers, and each Consul Data Center can have up to ~10,000 nodes and multiple Consul Data Centres can be connected over the WAN so that clients can discover nodes in other Data Centres. We recommend running a Consul Cluster in every Nomad Data Center and connecting them over the WAN. Please refer to the Consul documentation to learn more about bootstrapping Consul and connecting multiple Consul clusters over the WAN.
Also, Nomad clusters can be significantly larger than Consul clusters, so sharding the Consul clusters per ~10,000 nodes organized in individual DCs helps scale Consul as the Nomad clusters scale.
Bootstrapping a Nomad cluster without Consul
At least one Nomad server's address (also known as the seed node) needs to be
known ahead of time and a running agent can be joined to a cluster by running
the server-join cli command.
For example, once a Nomad agent starts in the server mode it can be joined to an existing cluster with a server whose IP is known. Once the agent joins the other node in the cluster, it can discover the other nodes via the gossip protocol.
nomad server-join -retry-join 10.0.0.1
The -retry-join parameter indicates that the agent should keep trying to join
the server even if the first attempt fails. This is essential when the other
address is going to be eventually available after some time as nodes might take
a variable amount of time to boot up in a cluster.
On the client side, the addresses of the servers are expected to be specified via the client configuration.
client {
...
servers = ["10.10.11.2:4648", "10.10.11.3:4648", "10.10.11.4:4648"]
...
}
In the above example we are specifying three servers for the clients to connect. If servers are added or removed, clients know about them via the heartbeat of a server which is alive.
Bootstrapping a Nomad cluster with Consul
Bootstrapping a Nomad cluster is significantly easier if Consul is used along with Nomad. If a local Consul cluster is bootstrapped before Nomad, the following configuration would register the Nomad agent with Consul and look up the addresses of the other Nomad server addresses and join with them automatically.
{
"server_service_name": "nomad-server",
"server_auto_join": true,
"client_service_name": "nomad-client",
"client_auto_join": true
}
With the above configuration Nomad agent is going to look up Consul for
addresses of agents in the nomad-server service and join them automatically.
In addition, if the auto-advertise option is set Nomad is going to register
the agents with Consul automatically too.
Please refer to the documentation for the complete set of configuration options.
Fedarating a cluster
Nomad clusters across multiple regions can be fedarated and once they are connected users can target the Nomad Servers in various regions from any other region while submitting a job or querying any Nomad API.
Fedarating multiple Nomad clusters is as simple as joining a server to any other remote server.
nomad server-join 10.10.11.8:4648
Servers across regions discover other servers in the cluster via the gossip protocol and hence it enough to join one known server.