»Upgrading

Nomad is designed to be flexible and resilient when upgrading from one Nomad version to the next. Upgrades should cause neither a Nomad nor a service outage. However, there are some restrictions to be aware of before upgrading:

Nomad strives to be backward compatible for at least 1 point release, so Nomad v0.10 hosts work with v0.9 hosts. Upgrading 2 point releases (eg v0.8 to v0.10) may work but is untested and unsupported.
- Nomad does not support downgrading at this time. Downgrading clients requires draining allocations and removing the data directory. Downgrading servers safely requires re-provisioning the cluster.
- New features are unlikely to work correctly until all nodes have been upgraded.
- Check the version upgrade details page for important changes and backward incompatibilities.
When upgrading a Nomad Client, if it takes longer than the heartbeat_grace (10s by default) period to restart, all allocations on that node may be rescheduled.

Nomad supports upgrading in place or by rolling in new servers:

In Place: The Nomad binary can be updated on existing hosts. Running allocations will continue running uninterrupted.
Rolling: New hosts containing the new Nomad version may be added followed by the removal of old hosts. The old nodes must be drained to migrate running allocations to the new nodes.

This guide describes both approaches.

Upgrade Process

Once you have checked the upgrade details for the new version, the upgrade process is as simple as updating the binary on each host and restarting the Nomad service.

At a high level we complete the following steps to upgrade Nomad:

Add the new version
Check cluster health
Remove the old version
Check cluster health
Upgrade clients

1. Add the new version to the existing cluster

While it is possible to upgrade Nomad client nodes before servers, this guide recommends upgrading servers first as many new client features will not work until servers are upgraded.

In a federated cluster, new features are not guaranteed to work until all agents in a region and the server nodes in the authoritative region are upgraded.

Whether you are replacing Nomad in place on existing systems or bringing up new servers you should make changes incrementally, verifying cluster health at each step of the upgrade.

On a single server, install the new version of Nomad. You can do this by joining a new server to the cluster or by replacing or upgrading the binary locally and restarting the Nomad service.

2. Check cluster health

Monitor the Nomad logs on the remaining servers to check that the new server has joined the cluster correctly.

Run nomad agent-info on the new servers and check that the last_log_index is of a similar value to the other servers. This step ensures that changes have been replicated to the new server.

ubuntu@nomad-server-10-1-1-4:~$ nomad agent-info
nomad
  bootstrap = false
  known_regions = 1
  leader = false
  server = true
raft
  applied_index = 53460
  commit_index = 53460
  fsm_pending = 0
  last_contact = 54.512216ms
  last_log_index = 53460
  last_log_term = 1
  last_snapshot_index = 49511
  last_snapshot_term = 1
  num_peers = 2
...

Continue with the upgrades across the servers making sure to do a single Nomad server at a time. You can check state of the servers with nomad server members, and the state of the client nodes with nomad node status.

3. Remove the old versions from servers

If you are doing an in place upgrade on existing servers this step is not necessary as the version was changed in place.

If you are doing an upgrade by adding new servers and removing old servers from the fleet you need to ensure that the server has left the fleet safely.

Stop the service on the existing host
On another server issue a nomad server members and check the status, if the server is now in a left state you are safe to continue.
If the server is not in a left state, issue a nomad server force-leave <server id> to remove the server from the cluster.

Monitor the logs of the other hosts in the Nomad cluster over this period.

4. Check cluster health

Use the same actions in step #2 above to confirm cluster health.

5. Upgrade clients

Following the successful upgrade of the servers you can now update your clients using a similar process as the servers. You may either upgrade clients in-place or start new nodes on the new version. See the Workload Migration Guide for instructions on how to migrate running allocations from the old nodes to the new nodes with the nomad node drain command.

Done

You are now running the latest Nomad version. You can verify all Clients joined by running nomad node status and checking all the clients are in a ready state.

Upgrading to Nomad Enterprise

The process of upgrading to a Nomad Enterprise version is identical to upgrading between versions of open source Nomad. The same guidance above should be followed and as always, prior to starting the upgrade please check the specific version details page as some version differences may require specific steps.

Upgrading to Raft Protocol 3

This section provides details on upgrading to Raft Protocol 3. Raft protocol version 3 requires Nomad running 0.8.0 or newer on all servers in order to work. Raft protocol version 2 will be removed in Nomad 1.4.0.

To see the version of the Raft protocol in use on each server, use the nomad operator raft list-peers command.

Note that the format of peers.json used for outage recovery is different when running with the latest Raft protocol. See Manual Recovery Using peers.json for a description of the required format.

When using Raft protocol version 3, servers are identified by their node-id instead of their IP address when Nomad makes changes to its internal Raft quorum configuration. This means that once a cluster has been upgraded with servers all running Raft protocol version 3, it will no longer allow servers running any older Raft protocol versions to be added.

Upgrading a Production Cluster to Raft Version 3

For production raft clusters with 3 or more members, the easiest way to upgrade servers is to have each server leave the cluster, upgrade its raft_protocol version in the server block (if upgrading to a version lower than v1.3.0), and then add it back. Make sure the new server joins successfully and that the cluster is stable before rolling the upgrade forward to the next server. It's also possible to stand up a new set of servers, and then slowly stand down each of the older servers in a similar fashion.

For in-place raft protocol upgrades, perform the following for each server, leaving the leader until last to reduce the chance of leader elections that will slow down the process:

Stop the server.
Run nomad server force-leave $server_name.
If the upgrade is for a Nomad version lower than v1.3.0, update the raft_protocol in the server's configuration file to 3.
Restart the server.
Run nomad operator raft list-peers to verify that the RaftProtocol for the server is now 3.
On the server, run nomad agent-info and check that the last_log_index is of a similar value to the other servers. This step ensures that raft is healthy and changes are replicating to the new server.

Upgrading a Single Server Cluster to Raft Version 3

If you are running a single Nomad server, restarting it in-place will result in that server not being able to elect itself as a leader. To avoid this, create a new raft.peers file before restarting the server with the new configuration. If you have jq installed you can run the following script on the server's host to write the correct raft.peers file:

#!/usr/bin/env bash

NOMAD_DATA_DIR=$(nomad agent-info -json | jq -r '.config.DataDir')
NOMAD_ADDR=$(nomad agent-info -json | jq -r '.stats.nomad.leader_addr')
NODE_ID=$(cat "$NOMAD_DATA_DIR/server/node-id")

cat <<EOF > "$NOMAD_DATA_DIR/server/raft/peers.json"
[
  {
    "id": "$NODE_ID",
    "address": "$NOMAD_ADDR",
    "non_voter": false
  }
]
EOF

After running this script, if the upgrade is for a Nomad version lower than v1.3.0, update the raft_protocol in the server's configuration to 3 and restart the server.