Kerek is a CLI that adds these missing pieces to Docker:
- Zero-downtime deployments for Docker Compose.
- Distributing container images via SSH instead of a registry.
- Using a custom SSH config file when connecting to a remote Docker instance.
- Docker or Podman
- Docker Compose
cargo install --git https://github.com/evolutics/kerek
cargo uninstall kerek
When updating services with docker compose up
, the old container of a service
is stopped before a new container of the service is started (stop-first
case):
Old container Stop
┄┄┄┄───────────────────────┤
Start New container
├────────────────────────┄┄┄┄
This causes a service interruption as there is a time when neither container is available.
Imagine that we could make the container lifetimes overlap instead
(start-first
case):
Old container Stop
┄┄┄┄───────────────────────────────────────┤
Start New container
├────────────────────────────────────────┄┄┄┄
Kerek supports both of these cases.
For services where a new container should be started before the old container is stopped, configure this in your Compose file:
# compose.yaml
services:
my-service:
image: my-image
+ deploy:
+ update_config:
+ order: start-first
Now just run kerek deploy
in place of docker compose up
. No need to change
any other Docker Compose workflows.
Above config is an optional part of the
Compose specification.
But vanilla Docker Compose always updates services in stop-first
order, even
if your Compose file says otherwise.
There is a demo with an
example service called greet
, which should stay available during updates.
For that purpose, traffic on localhost:8080 is proxied trough a second service
reverse-proxy
(Caddy):
localhost:8080 ╭───────────────╮ ╭───────────────╮
──────────────────────┤ reverse-proxy ├────────┤ greet │
:81 │ (stop-first) │ :80 │ (start-first) │
╰───────────────╯ ╰───────────────╯
With greet
configured in start-first
update order, the reverse-proxy
can
seamlessly switch over traffic from old to new version of the proxied service.
Try it yourself:
cd examples/zero_downtime_deployment
kerek deploy --wait
curl localhost:8080 # … prints "Hi from A"
export GREET_VERSION=B
kerek deploy --wait
curl localhost:8080 # … prints "Hi from B"
docker compose down
To see above deployments in action, use a separate shell session to run
while true; do curl --fail --max-time 0.2 localhost:8080; sleep 0.01s; done
There is an example of a continuous delivery pipeline with these essential steps:
- Provision a Vagrant VM with Podman, which serves as a staging environment.
- Temporarily run an SSH tunnel to the Podman API service on staging.
- Transfer container images to staging via this SSH tunnel.
- Deploy the Compose application to staging via the SSH tunnel.
- Do a smoke test.
Pass --container-engine podman
or set the environment variable
CONTAINER_ENGINE=podman
to use Podman instead of Docker. Kerek is
engine-agnostic so you may use any other container engine with a compatible CLI.
Podman Compose is not supported as it currently lacks some needed features like
the calculation of service config hashes (docker compose config --hash \*
).
Kerek may be set up as a Docker CLI plugin. With that, calls to subcommands like
kerek deploy
can be replaced by docker deploy
, which some people prefer.
Example setup:
mkdir --parents ~/.docker/cli-plugins
ln --symbolic ~/.cargo/bin/kerek ~/.docker/cli-plugins/docker-deploy
docker deploy --help
Other lightweight options for continuous delivery of containerized applications:
- Docker without Compose
docker rollout
- Docker Swarm mode
- K3s
- PaaS like CapRover, Dokku, etc.
podman kube play
Lightweight continuous delivery for Docker Compose
Usage: kerek [OPTIONS] <COMMAND>
Commands:
deploy Create or update services
provision Provisions host with container engine, making system-wide
changes
transfer-images Copies images from default to specified Docker host
tunnel-ssh Forwards local Unix domain socket to remote Docker host over
SSH
help Print this message or the help of the given subcommand(s)
Options:
--container-engine <CONTAINER_ENGINE>
Container engine program to use
[env: CONTAINER_ENGINE=]
[default: docker]
--dry-run
Do not apply changes, only show what would be done
--config <CONFIG>
Location of client config files
-c, --context <CONTEXT>
Name of the context to use to connect to the daemon (overrides
DOCKER_HOST env var and default context set with `docker context use`)
-D, --debug
Enable debug mode
-H, --host <HOST>
Daemon socket to connect to
-l, --log-level <LOG_LEVEL>
Set the logging level
[possible values: debug, info, warn, error, fatal]
--tls
Use TLS; implied by --tlsverify
--tlscacert <TLSCACERT>
Trust certs signed only by this CA
--tlscert <TLSCERT>
Path to TLS certificate file
--tlskey <TLSKEY>
Path to TLS key file
--tlsverify
Use TLS and verify the remote
-h, --help
Print help
-V, --version
Print version
Create or update services
Builds, (re)creates, and starts containers for a service.
If service names are given as command-line operands, this command does not
automatically start any of their linked services.
The containers are always run in the background (detached mode).
If there are existing containers for a service whose service config has changed
since the containers' creation, the changes are applied by recreating the
containers (preserving mounted volumes).
More precisely, a service is updated only if its service config hash changes
(details in https://github.com/docker/compose/blob/main/pkg/compose/hash.go).
Note that the service config hash does not depend on the container image
contents but only the `image` field. Thus, reusing an image tag like `latest`
does not trigger an update.
To force updating services regardless of config hash changes, use the
`--force-recreate` flag.
In summary:
| Command | Effect |
| --------------------------------- | -------------------------------------- |
| `kerek deploy` | Update services with changed hash |
| `kerek deploy x` | Update service `x` if its hash changed |
| `kerek deploy --force-recreate` | Always update all services |
| `kerek deploy --force-recreate x` | Always update service `x` |
| `kerek deploy --dry-run` | Only show what would be changed |
| `docker compose config --hash \*` | Show service config hashes |
Whether the old containers are stopped before or after the new containers are
started is controlled via `services.*.deploy.update_config.order` in a Compose
file. The options are `stop-first` and `start-first`, respectively.
Services are updated in lexicographical order (by Unicode code point),
regardless of dependencies. For each service, containers are stopped then
started (`stop-first`, default) or started then stopped (`start-first`),
respectively, and this is repeated for replicas:
- `stop-first` case:
1. Stop old replica 1
2. Start new replica 1
3. Stop old replica 2
4. Start new replica 2
5. …
- `start-first` case:
1. Start new replica 1
2. Stop old replica 1
3. Start new replica 2
4. Stop old replica 2
5. …
Usage: kerek deploy [OPTIONS] [SERVICE_NAMES]...
Arguments:
[SERVICE_NAMES]...
Services to consider
Options:
--all-resources
Include all resources, even those not used by services
--ansi <ANSI>
Control when to print ANSI control characters
[possible values: never, always, auto]
--compatibility
Run compose in backward compatibility mode
--env-file <ENV_FILE>
Specify an alternate environment file
-f, --file <FILE>
Compose configuration files
--parallel <PARALLEL>
Control max parallelism, -1 for unlimited
--profile <PROFILE>
Specify a profile to enable
--progress <PROGRESS>
Set type of progress output
[possible values: auto, tty, plain, json, quiet]
--project-directory <PROJECT_DIRECTORY>
Specify an alternate working directory (default: the path of the,
first specified, Compose file)
-p, --project-name <PROJECT_NAME>
Project name
--build
Build images before starting containers
--force-recreate
Recreate containers even if their configuration and image haven't
changed
--no-build
Don't build an image, even if it's policy
--no-start
Don't start the services after creating them
--pull <PULL>
Pull image before running
[possible values: always, missing, never]
--quiet-pull
Pull without printing progress information
--remove-orphans
Remove containers for services not defined in the Compose file
-V, --renew-anon-volumes
Recreate anonymous volumes instead of retrieving data from the
previous containers
-t, --timeout <TIMEOUT>
Use this timeout in seconds for container shutdown when containers are
already running
--wait
Wait for services to be running|healthy
--wait-timeout <WAIT_TIMEOUT>
Maximum duration to wait for the project to be running|healthy
-h, --help
Print help (see a summary with '-h')
Provisions host with container engine, making system-wide changes
This targets a host via SSH, unless host `localhost` and no SSH config file are
passed as arguments, in which case the current machine is targeted.
Usage: kerek provision [OPTIONS] <HOST>
Arguments:
<HOST>
Reference like `localhost` or `[ssh://][<user>@]<hostname>[:<port>]`
Options:
--force
Go ahead without prompting user to confirm
-F, --ssh-config <SSH_CONFIG>
Path to SSH config file
-h, --help
Print help (see a summary with '-h')
Copies images from default to specified Docker host
By default, only images absent on the destination host are transferred. An image
is considered present if the name matches one of these forms:
- `<namespace>:<tag>`
- `<namespace>@<digest>`
- `<namespace>:<tag>@<digest>`
Examples:
kerek --host ssh://192.0.2.1 transfer-images my-img
DOCKER_HOST=ssh://from kerek --host ssh://to transfer-images my-img
DOCKER_CONTEXT=from kerek --context to transfer-images my-img
docker compose config --images | kerek --host … transfer-images -
kerek --host … transfer-images --compress zstd my-img
kerek --host … transfer-images --compress 'xz -9' my-img
Usage: kerek transfer-images [OPTIONS] [IMAGES]...
Arguments:
[IMAGES]...
Images to copy; use `-` to pass image names as stdin lines
Options:
--compress <COMPRESS>
Compression command to use (`bzip2`, `gzip`, `xz`, `zstd`, etc.)
--force
Copy images without checking if the destination already has such
images; useful for replacing images with `latest` tag
-h, --help
Print help (see a summary with '-h')
Forwards local Unix domain socket to remote Docker host over SSH
This runs an SSH tunnel in the background. Meanwhile, you can connect to the
remote Docker host using `DOCKER_HOST=unix:///path/to/kerek.sock` locally. Note
that a custom SSH config file can be specified, unlike with vanilla Docker.
Example:
kerek tunnel-ssh my-ssh-host
CONTAINER_HOST="unix://${PWD}/kerek.sock" podman ps
fuser --kill -TERM kerek.sock
Usage: kerek tunnel-ssh [OPTIONS] <SSH_HOST>
Arguments:
<SSH_HOST>
Reference like `[ssh://][<user>@]<hostname>[:<port>]`
Options:
--local-socket <LOCAL_SOCKET>
Path to Unix domain socket on localhost to be forwarded
[default: kerek.sock]
--remote-socket <REMOTE_SOCKET>
Path to Unix domain socket of Docker host on remote
-F, --ssh-config <SSH_CONFIG>
Path to SSH config file
-h, --help
Print help (see a summary with '-h')