Frequently Asked Questions

Q: What is Checkpoint? / Does Nomad call home?

Nomad makes use of a HashiCorp service called Checkpoint which is used to check for updates and critical security bulletins. Only anonymous information, which cannot be used to identify the user or host, is sent to Checkpoint. An anonymous ID is sent which helps de-duplicate warning messages. This anonymous ID can be disabled. Using the Checkpoint service is optional and can be disabled.

See disable_anonymous_signature and disable_update_check.

Q: Is Nomad eventually or strongly consistent?

Nomad makes use of both a consensus protocol and a gossip protocol. The consensus protocol is strongly consistent, and is used for all state replication and scheduling. The gossip protocol is used to manage the addresses of servers for automatic clustering and multi-region federation. This means all data that is managed by Nomad is strongly consistent.

Q: Is Nomad's datacenter parameter the same as Consul's?

No. For those familiar with Consul, Consul's notion of a datacenter is more equivalent to a Nomad region. Nomad supports grouping nodes into multiple datacenters, which should reflect nodes being colocated, while being managed by a single set of Nomad servers.

Consul on the other hand does not have this two-tier approach to servers and agents and instead relies on federation to create larger logical clusters.

Q: What is "bootstrapping" a Nomad cluster?

Bootstrapping is the process when a Nomad cluster elects its first leader and writes the initial cluster state to that leader's state store. Bootstrapping will not occur until at least a given number of servers, defined by bootstrap_expect, have connected to each other. Once this process has completed, the cluster is said to be bootstrapped and is ready to use.

Certain configuration options are only used to influence the creation of the initial cluster state during bootstrapping and are not consulted again so long as the state data remains intact. These typically are values that must be consistent across server members. For example, the default_scheduler_config option allows an operator to set the SchedulerConfig to non-default values during this bootstrap process rather than requiring an immediate call to the API once the cluster is up and running.

If the state is completely destroyed, whether intentionally or accidentally, on all of the Nomad servers in the same outage, the cluster will re-bootstrap based on the Nomad defaults and any configuration present that impacts the bootstrap process.