Skip to main content

v0.22.0

This page mirrors the GitHub release, which also includes the full list of commits.

This release makes cluster storage for Caddy the default, adds service and container labels, a new uc service update command to change a service without recreating its containers, uc image rm to clean up images across machines, and DNS names for machines. It also brings fixes for logs, faster stop-first deployments, and a much calmer daemon on busy clusters.

⚠️ Heads-up before upgrading​

  • uc caddy deploy now deploys the Uncloud Caddy image with cluster storage (22b1c68). Upgrading Uncloud doesn't touch your running Caddy. But the next uc caddy deploy switches it to the new image and storage. Caddy doesn't copy certificates between storage backends, so it will reissue them unless you migrate them first. Read Upgrading from earlier versions before you redeploy Caddy.
  • uc caddy deploy requires the 0.22.0 daemon on all machines. Upgrade the daemons first.
  • Machine names must be valid DNS labels (ce1c6bd). This means lowercase letters, numbers, and hyphens, up to 63 characters. The names rr and nearest are reserved, and a name can't look like a machine ID. This applies when you add or rename a machine.

Cluster storage for Caddy by default​

Change: 22b1c68, docs: Managing Caddy

In v0.21.0, we added an opt-in cluster storage for Caddy to share TLS certificates across machines. Now it's the default.

New clusters and uc caddy deploy now use the Uncloud Caddy image ghcr.io/unlabs-dev/caddy-uncloud. It's a Caddy build with the Uncloud storage module and a few popular modules, such as caddy-l4 and cloudflare for DNS challenges. So you can get a wildcard certificate with Cloudflare without building your own image.

uc caddy deploy also remembers the --image, --caddyfile, and --machine values you passed. You no longer need to repeat them every time you upgrade Caddy. Pass an empty value to reset a flag to its default:

# Deploy with a custom global config only on two machines.
uc caddy deploy --caddyfile global.Caddyfile --machine machine1,machine2

# Later, upgrade Caddy to the latest image. The custom config and machines are kept.
uc caddy deploy

# Go back to the default global config and deploy to all machines.
uc caddy deploy --caddyfile "" --machine ""

If you have an existing cluster, follow Upgrading from earlier versions. It explains how to keep your certificates when switching to cluster storage, including for Caddy managed with Compose.

Service and container labels​

Changes: 81f30aa, c1a7f84, 7b48ac5

You can now set labels in Compose. Uncloud supports both kinds of labels:

  • labels sets Docker labels on containers, for tools like backup agents, log shippers, or monitoring agents.
  • deploy.labels sets service labels. They're stored with the service but not applied to containers. Use them for metadata about the service itself. Uncloud already uses them, for example, to mark the Caddy service managed by uc caddy deploy.
services:
  db:
    image: postgres:18
    labels:
      docker-volume-backup.stop-during-backup: "true"
    deploy:
      labels:
        owner: [email protected]

uc run gets the matching -l/--label and --container-label flags. And uc inspect shows the labels of a service.

The deployment plan shows exactly which labels change. Service labels are updated in place, so changing them doesn't restart your containers:

~ update service web
  │ ~ labels:
  │     + env=prod
  │     ~ [email protected] → [email protected]
  ╰── ~   update container web/0f6df6593e1b on machine-1 (in place)

Changing container labels still recreates containers because Docker can't change labels on an existing container.

Update services in place​

Changes: ff6ac45, 81f30aa

Uncloud now updates CPU and memory limits on running containers instead of recreating them. This works with uc deploy and the new uc service update command:

# Change the CPU and memory limits.
uc service update web --cpu 0.5 --memory 512m

# Add or change a service label and remove another one.
uc service update web --label-add env=prod --label-rm owner

This is handy when a service is running out of memory and you want to bump the previously set limit right away without a restart.

uc run and uc service update also have a new --memory-reservation flag to set the minimum amount of memory reserved for a container.

Remove images from machines​

Changes: e9eab32, 7c6d1be, 7e8736f.

Old images pile up quickly when you push and deploy often. The new uc image rm command removes them from all or selected machines. It shows a removal plan and asks for confirmation first:

# Remove a tag from all machines.
uc image rm myapp:old

# Remove all tags of an image.
uc image rm myapp

# Remove matching tags from two machines without confirmation.
uc image rm 'myapp:1.*' 'otherapp:*' -m machine1,machine2 --yes

Don't worry about breaking anything. Images used by running or stopped containers can't be removed.

uc image ls now lists all tags and digest references of each image too, not just the first tag.

DNS names for machines​

PR: #359, issue: #340. Thanks to @miekg for the contribution ❤️

Containers can now reach machines by name or ID in the internal m.internal zone. This is handy when a container needs to reach something running directly on a machine outside of Docker.

$ dig +short machine-1.m.internal
10.210.0.1

A query for m.internal returns the IPs of all machines. See Internal DNS for more details.

Improvements​

  • Debounce container change notifications in the daemon. On busy clusters, every health check or container change triggered a full Caddy and DNS config reconcile on every machine. This could burn a lot of CPU and even freeze small machines (#438). Thanks to @dasunsrule32 for the contribution ❤️
  • Create the new container and pull its image before stopping the old one in stop-first deployments. This shortens downtime during the cutover (#420). Thanks to @nick-potts for the contribution ❤️
  • --since and --until log filters now support days, for example, uc logs web --since 2d (5e26cfb).
  • uc context show now respects the UNCLOUD_CONTEXT environment variable (#444). Thanks to @miekg for the contribution ❤️

Bug fixes​

  • Resolve secret:// references in pre-deploy hook environment and don't replicate their values to the cluster store, same as for service environment (#424, #422). Thanks to @vky5 for the contribution ❤️
  • Show all matching logs when --since is set instead of only the last 100 lines per replica (409dc3b).
  • Use your local time zone when resolving dates and times in --since and --until log filters (5fa2368).
  • Disable colors in logs and uc caddy config output when redirected to a file or when NO_COLOR is set (a1b6b31, 80d2a32).
  • Don't count stopped containers as removed in the deployment plan summary (6c7ddeb).
  • Fix the default protocol for published ports in the uc run help. It's tcp, not http (99c928f).

Upgrade to v0.22.0​

Upgrade your local CLI and the daemon on every machine. If you're upgrading from v0.20.x or earlier, follow the v0.21.0 upgrade instructions first for the systemd unit changes.

Uncloud CLI locally​

# Homebrew (macOS, Linux)
brew upgrade uncloud

# Install script (macOS, Linux)
curl -fsS https://get.uncloud.run/install.sh | VERSION=0.22.0 sh

Machine daemon​

Run the following commands over SSH on each machine to upgrade the daemon:

ARCH=$(uname -m | sed 's/x86_64/amd64/;s/aarch64/arm64/')
curl -fsSL -o uncloudd.tar.gz https://github.com/psviderski/uncloud/releases/download/v0.22.0/uncloudd_linux_${ARCH}.tar.gz
tar -xf uncloudd.tar.gz
sudo install uncloudd /usr/local/bin/uncloudd
rm uncloudd uncloudd.tar.gz
sudo systemctl restart uncloud

After upgrading​

Check your CLI and daemon versions:

uc version
uc machine ls

Then switch Caddy to the new image and cluster storage when you're ready. Follow Upgrading from earlier versions to keep your certificates.