On a single host¶
Usually you would run
deck-chores in a container:
$ docker run --rm -v /var/run/docker.sock:/var/run/docker.sock funkyfuture/deck-chores:1
There’s a manifest on the Docker Hub that maps images to builds targeting
Thus you don’t need to specify any platform indicator, the Docker client will figure out which
one is the proper image to pull.
Likewise, docker-compose can be used with such configuration:
version: "3.7" services: officer: image: funkyfuture/deck-chores:1 restart: unless-stopped environment: TIMEZONE: Asia/Tel Aviv volumes: - /var/run/docker.sock:/var/run/docker.sock
You could also install
deck-chores from the Python Package Index with
$ pipx install deck-chores
and then run it:
Now one instance of
deck-chores is running and will handle all job definitions that it discovers
on containers that run on the Docker host.
In a Docker Swarm¶
deck-chores can be run in a Docker Swarm cluster, but it must be deployed on all nodes and it
cannot restrict jobs to be run in only one of the containers that manifest a service. This would be
a suitable stack definition:
version: "3.7" services: officer: image: funkyfuture/deck-chores:1 deploy: mode: global environment: TIMEZONE: Europe/Berlin # it isn't guaranteed that service or job options don't override this: DEFAULT_FLAGS: noservice volumes: - /var/run/docker.sock:/var/run/docker.sock
It can be deployed with:
$ docker stack deploy --compose-file docker-compose.yml deck-chores
Now one instance of
deck-chores is running on each Swarm node and each will handle all job
definitions that it discovers on containers that run on the same Swarm node. No instance is aware
of the events and containers on other nodes.
Caveats & Tips¶
There’s yet no way to distinguish container events that happen during an image build from others (#6 and #15211). Thus when an image is built, deck-chores will register and remove jobs on all intermediate containers following labels that define jobs. It would possibly trigger these jobs, which might lead to a corrupted build. You can avoid this risk by building images on a host that is not observed by deck-chores or by pausing it during image builds. Another alternative could be using Podman to build images.
Containers without an enduring main process¶
If the container is supposed to only run the scheduled commands and not a main process, use a
non-stopping no-op command as main process like in this snippet of a
services: neverending: # … command: tail -f /dev/null labels: deck-chores.daily_job.command: daily_command … deck-chores.daily_job.interval: daily
Making jobs’ output available to
docker logs of the executing container¶
Docker captures the output of the first process in a container as logged data. In order to capture
the output of a job’s command as well, its output needs to be redirected to the main process’
stderr, e.g. with by redirecting a command’s output with a shell:
deck-chores.a_job.command: sh -c "/usr/local/bin/job_script.sh &> /proc/1/fd/1"
Listing all registered jobs¶
Information, including the next scheduled execution, about the registered jobs of a deck-chores
instance can be logged at once by sending
SIGUSR1 signal to the process, e.g. to one that runs
in a container:
docker kill --signal USR1 <CONTAINER>
Job definitions are parsed from a container’s metadata aka labels. A label’s key must be in the
namespace defined by
deck-chores) to be considered. A job
has its own namespace that holds all its attributes. Thus an attribute’s key has usually this
$LABEL_NAMESPACE.<job name>.<job attribute>
An exception is a job’s
env namespace that is structured like this:
$LABEL_NAMESPACE.<job name>.env.<variable name>
The job name
options cannot be used as it is reserved for setting Container-scoped configuration.
A job name can consist of lower-case letters, digits and dashes.
The following attributes are available:
|command||the command to run|
|cron||a cron definition|
|date||a date definition|
|env||this namespace holds environment variables that are set on the command’s execution context|
|interval||an interval definition|
|jitter||the maximum length of a random delay before each job’s execution (in conjunction with a cron or interval trigger); can be either a number that define seconds or a number with a subsequent time unit indicator like the interval trigger|
|max||the maximum of simultaneously running command instances, defaults to
|timezone||the timezone that the trigger relates to, defaults to
|user||the user to run the command; see the user option for details regarding the defaults|
|workdir||the working directory when the command is executed|
command and one of
interval are required for each
Example snippet from a
services: web: # ... labels: deck-chores.clear-caches.command: drush cc all deck-chores.clear-caches.interval: daily deck-chores.clear-caches.user: www-data deck-chores.clear-caches.env.ENVIRONMENT: production
Or baked into an image:
LABEL deck-chores.clear-caches.command="drush cc all" \ deck-chores.clear-caches.interval="daily" \ deck-chores.clear-caches.user="www-data" \ deck-chores.clear-caches.env.ENVIRONMENT="production"
cron triggers allow definitions for repeated run times like for the well-known cron daemon.
In contrast to the classic, the sequence of fields is flipped, starting with the greatest unit
on the left. The fields are separated by spaces, missing fields are filled up with
* on the
The fields from left to right define:
See APScheduler’s documentation for details on its versatile expressions.
* * * * * */3 0 0 # run on all hours dividable by 3 */3 0 0 # as shortened expression * * * * 6 1 0 0 # run every Sunday at 1:00 6 1 0 0 # as shortened expression sun 1 0 0 # as 'speaking' variant * * * * * 1-4 0 0 # run daily at 1:00, 2:00, 3:00 and 4:00 1-4 0 0 # as shortened expression
A one-time trigger that is formatted as
An omitted time is interpreted as
0:00:00. Note that times must include a seconds field.
This trigger defines a repetition by a fixed interval. It can either be a string where time units follow numbers or a sequence of numbers that qualify time units by order.
In the first form the numbers can be decimal fractions and the time units are determined by the first letter of a token as week, day, hour, minute or second.
In the anonymous form the interval is added up by the fields weeks, days, hours, minutes
and seconds in that order. Possible field separators are
/ and spaces. Missing
fields are filled up with
0 on the left.
28 Days # run every 4 weeks 4 wookies # run every 4 weeks 42s 0.5d # run every twelve hours and 42 seconds 42:00:00 # run every fourty-two hours 100/00:00:00 # run every one hundred days
There are also the convenience shortcuts
every minute and
Though it uses the same units of measurement, an interval is different from a recurring point in time of a specific calendar system, it describes the time between two events. Hence you should expect a job that is defined with this type of trigger to run the defined time after the job has been registered. To define a recurring point in time, see the cron trigger.
deck-chores doesn’t track jobs’ status when they are removed from the scheduler
and doesn’t persist any data between its invocations. Any such event would therefore reset the
next scheduled run time of a job. Depending on a system’s usage this is more or less likely
to happen. For longer intervals, a cron trigger would therefore be preferable.
A user that shall run all jobs for a container can be set with a label name of this form:
The option can also be defined for an image and is considered when the
flag is set.
If this option is not set, Docker uses the user that was specified with the
--user option on
container creation or falls back to the one defined in the underlying image.
Option flags control deck-chores’s behaviour with regard to the labeled container and override
the setting of
DEFAULT_FLAGS. The schema for a flags label name is:
Options are set as comma-separated list of flags. An option set by
be unset by prefixing with
These options are available:
Job definitions in the container’s basing image labels are also parsed while container label keys override these.
deck-chore’s behaviour is defined by these environment variables:
The timeout for responses from the Docker daemon in seconds without unit indicator. The default is imported from docker-py.
The size of caches that save immutable container properties, like the parsed and possibly absent job definitions. Since memory is cheap and so are the stored objects, increase this when you have a lot of containers floating around to reduce latency.
The URL of the Docker daemon to connect to.
Log debugging messages, enabled by
The default for a job’s
The pool size of job executors defines the maximum number of jobs that can run at the same time.
The label namespace to look for job definitions and container options.
A comma-separated list of container labels that identify a unique service with possibly multiple container instances. This has an impact on how the
The job scheduler’s timezone and the default for a job’s
TLS(selects the highest version supported by the client and the daemon)
For other options see the names provided by Python’s ssl library prefixed with
Authentication related files are expected to be available at