Autopilot

Autopilot is a set of new features added in Nomad 0.8 to allow for automatic operator-friendly management of Nomad servers. It includes cleanup of dead servers, monitoring the state of the Raft cluster, and stable server introduction.

To enable Autopilot features (with the exception of dead server cleanup), the raft_protocol setting in the server stanza must be set to 3 on all servers. This setting defaults to 2; a cluster configured with protocol 2 can be upgraded to protocol 3 with a rolling update, provided time for membership to stabilize following each server update. During an upgrade from raft protocol 2 to 3, use the nomad operator raft list-peers command between server updates to verify that each server identifier is replaced with a UUID. For more information, consult the Version Upgrade section on Raft Protocol versions.

Configuration

The configuration of Autopilot is loaded by the leader from the agent's Autopilot settings when initially bootstrapping the cluster:

autopilot {
  cleanup_dead_servers      = true
  last_contact_threshold    = "200ms"
  max_trailing_logs         = 250
  server_stabilization_time = "10s"
  enable_redundancy_zones   = false
  disable_upgrade_migration = false
  enable_custom_upgrades    = false
}

After bootstrapping, the configuration can be viewed or modified either via the operator autopilot subcommand or the /v1/operator/autopilot/configuration HTTP endpoint:

View the configuration.

$ nomad operator autopilot get-config
CleanupDeadServers = true
LastContactThreshold = 200ms
MaxTrailingLogs = 250
ServerStabilizationTime = 10s
EnableRedundancyZones = false
DisableUpgradeMigration = false
EnableCustomUpgrades = false

Update the configuration.

$ nomad operator autopilot set-config -cleanup-dead-servers=false
Configuration updated!

View the configuration to confirm your changes.

$ nomad operator autopilot get-config
CleanupDeadServers = false
LastContactThreshold = 200ms
MaxTrailingLogs = 250
ServerStabilizationTime = 10s
EnableRedundancyZones = false
DisableUpgradeMigration = false
EnableCustomUpgrades = false

Dead server cleanup

Dead servers will periodically be cleaned up and removed from the Raft peer set, to prevent them from interfering with the quorum size and leader elections. This cleanup will also happen whenever a new server is successfully added to the cluster.

Prior to Autopilot, it would take 72 hours for dead servers to be automatically reaped, or operators had to script a nomad force-leave. If another server failure occurred, it could jeopardize the quorum, even if the failed Nomad server had been automatically replaced. Autopilot helps prevent these kinds of outages by quickly removing failed servers as soon as a replacement Nomad server comes online. When servers are removed by the cleanup process they will enter the "left" state.

This option can be disabled by running nomad operator autopilot set-config with the -cleanup-dead-servers=false option.

Server health checking

An internal health check runs on the leader to monitor the stability of servers. A server is considered healthy if all of the following conditions are true:

• Its status according to Serf is 'Alive'

• The time since its last contact with the current leader is below LastContactThreshold

• Its latest Raft term matches the leader's term

• The number of Raft log entries it trails the leader by does not exceed MaxTrailingLogs

The status of these health checks can be viewed through the /v1/operator/autopilot/health HTTP endpoint, with a top level Healthy field indicating the overall status of the cluster:

$ curl localhost:4646/v1/operator/autopilot/health
{
  "Healthy": true,
  "FailureTolerance": 0,
  "Servers": [
    {
      "ID": "e349749b-3303-3ddf-959c-b5885a0e1f6e",
      "Name": "node1",
      "Address": "127.0.0.1:4647",
      "SerfStatus": "alive",
      "Version": "0.8.0",
      "Leader": true,
      "LastContact": "0s",
      "LastTerm": 2,
      "LastIndex": 10,
      "Healthy": true,
      "Voter": true,
      "StableSince": "2017-03-28T18:28:52Z"
    },
    {
      "ID": "e35bde83-4e9c-434f-a6ef-453f44ee21ea",
      "Name": "node2",
      "Address": "127.0.0.1:4747",
      "SerfStatus": "alive",
      "Version": "0.8.0",
      "Leader": false,
      "LastContact": "35.371007ms",
      "LastTerm": 2,
      "LastIndex": 10,
      "Healthy": true,
      "Voter": false,
      "StableSince": "2017-03-28T18:29:10Z"
    }
  ]
}

Stable server introduction

When a new server is added to the cluster, there is a waiting period where it must be healthy and stable for a certain amount of time before being promoted to a full, voting member. This can be configured via the ServerStabilizationTime setting.

Server read and scheduling scaling

With the non_voting_server option, a server can be explicitly marked as a non-voter and will never be promoted to a voting member. This can be useful when more read scaling is needed; being a non-voter means that the server will still have data replicated to it, but it will not be part of the quorum that the leader must wait for before committing log entries. Non voting servers can also act as scheduling workers to increase scheduling throughput in large clusters.

Redundancy zones

Prior to Autopilot, it was difficult to deploy servers in a way that took advantage of isolated failure domains such as AWS Availability Zones; users would be forced to either have an overly-large quorum (2-3 nodes per AZ) or give up redundancy within an AZ by deploying only one server in each.

If the EnableRedundancyZones setting is set, Nomad will use its value to look for a zone in each server's specified redundancy_zone field.

Here's an example showing how to configure this:

/* config.hcl */
server {
  redundancy_zone = "west-1"
}

$ nomad operator autopilot set-config -enable-redundancy-zones=true
Configuration updated!

Nomad will then use these values to partition the servers by redundancy zone, and will aim to keep one voting server per zone. Extra servers in each zone will stay as non-voters on standby to be promoted if the active voter leaves or dies.

Upgrade migrations

Autopilot in Nomad Enterprise supports upgrade migrations by default. To disable this functionality, set DisableUpgradeMigration to true.

When a new server is added and Autopilot detects that its Nomad version is newer than that of the existing servers, Autopilot will avoid promoting the new server until enough newer-versioned servers have been added to the cluster. When the count of new servers equals or exceeds that of the old servers, Autopilot will begin promoting the new servers to voters and demoting the old servers. After this is finished, the old servers can be safely removed from the cluster.

To check the Nomad version of the servers, either the autopilot health endpoint or the nomad memberscommand can be used:

$ nomad server members
Name   Address    Port  Status  Leader  Protocol  Build  Datacenter  Region
node1  127.0.0.1  4648  alive   true    3         0.7.1  dc1         global
node2  127.0.0.1  4748  alive   false   3         0.7.1  dc1         global
node3  127.0.0.1  4848  alive   false   3         0.7.1  dc1         global
node4  127.0.0.1  4948  alive   false   3         0.8.0  dc1         global

Migrations without a Nomad version change

The EnableCustomUpgrades field can be used to override the version information used during a migration, so that the migration logic can be used for updating the cluster when changing configuration.

If the EnableCustomUpgrades setting is set to true, Nomad will use its value to look for a version in each server's specified upgrade_version tag. The upgrade logic will follow semantic versioning and the upgrade_version must be in the form of either X, X.Y, or X.Y.Z.