just-containers/s6-overlay

Table of Contents

• Quickstart
• Compatibility with v2
• Goals
• Features
• The Docker Way?
• Init stages
• Installation
• Usage
  • Using CMD
  • Writing a service script
  • Setting the exit code of the container to the exit code of your main service
  • Fixing ownership and permissions
  • Executing initialization and finalization tasks
  • Writing an optional finish script
  • Logging
  • Dropping privileges
  • Read-only Root Filesystem
  • Container environment
  • Customizing s6-overlay's behaviour
  • syslog
• Performance
• Verifying Downloads
• Notes
• Releases
  • Which architecture to use depending on your TARGETARCH
• Contributing
• Building the overlay yourself
• Upgrade notes

s6-overlay

s6-overlay is an easy-to-install (just extract a tarball or two!) set of scripts and utilities allowing you to use existing Docker images while using s6 as a pid 1 for your container and process supervisor for your services.

Quickstart

Build the following Dockerfile and try it out:

# Use your favorite image
FROM ubuntu
ARG S6_OVERLAY_VERSION=3.2.3.2

RUN apt-get update && apt-get install -y nginx xz-utils
RUN echo "daemon off;" >> /etc/nginx/nginx.conf
CMD ["/usr/sbin/nginx"]

ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-noarch.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-noarch.tar.xz
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-x86_64.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-x86_64.tar.xz
ENTRYPOINT ["/init"]

docker-host $ docker build -t demo .
docker-host $ docker run --name s6demo -d -p 80:80 demo
docker-host $ docker top s6demo acxf
PID                 TTY                 STAT                TIME                COMMAND
11735               ?                   Ss                  0:00                \_ s6-svscan
11772               ?                   S                   0:00                \_ s6-supervise
11773               ?                   Ss                  0:00                | \_ s6-linux-init-s
11771               ?                   Ss                  0:00                \_ rc.init
11812               ?                   S                   0:00                | \_ nginx
11814               ?                   S                   0:00                | \_ nginx
11816               ?                   S                   0:00                | \_ nginx
11813               ?                   S                   0:00                | \_ nginx
11815               ?                   S                   0:00                | \_ nginx
11779               ?                   S                   0:00                \_ s6-supervise
11785               ?                   Ss                  0:00                | \_ s6-ipcserverd
11778               ?                   S                   0:00                \_ s6-supervise
docker-host $ curl --head http://127.0.0.1/
HTTP/1.1 200 OK
Server: nginx/1.18.0 (Ubuntu)
Date: Mon, 17 Jan 2022 13:33:58 GMT
Content-Type: text/html
Content-Length: 612
Last-Modified: Mon, 17 Jan 2022 13:32:11 GMT
Connection: keep-alive
ETag: "61e56fdb-264"
Accept-Ranges: bytes

Compatibility with v2

If you're migrating from a previous version of s6-overlay (v2) to the new version (v3), you may need to make some changes to your services or the way you use s6-overlay in order for everything to work smoothly. This document tries to be accurate on how v3 works, but we have a separate page listing the main differences, and things you're likely to notice. Please read it if you're in this situation!

Goals

The project has the following goals:

• Be usable on top of any Docker image
• Make it easy to create new images, that will operate like any other images
• Provide users with a turnkey s6 installation that will give them a stable pid 1, a fast and orderly init sequence and shutdown sequence, and the power of process supervision and automatically rotated logs.

Features

• A simple init process which allows the end-user to execute tasks like initialization (cont-init.d), finalization (cont-finish.d) and their own services with dependencies between them
• The s6-overlay provides proper PID 1 functionality
  • You'll never have zombie processes hanging around in your container, they will be properly cleaned up.
• Multiple processes in a single container
• Able to operate in "The Docker Way"
• Usable with all base images - Ubuntu, CentOS, Fedora, Alpine, Busybox...
• Distributed as a small number of .tar.xz files depending on what exact functionality you need - to keep your image's number of layers small.
• A whole set of utilities included in s6 and s6-portable-utils. They include handy and composable utilities which make our lives much, much easier.
• Log rotating out-of-the-box through logutil-service which uses s6-log under the hood.
• Some support for Docker's USER directive, to run your whole process tree as a specific user. Not compatible with all features, details in the notes section.

The Docker Way?

One of the oft-repeated Docker mantras is "one process per container", but we disagree. There's nothing inherently bad about running multiple processes in a container. The more abstract "one thing per container" is our policy - a container should do one thing, such as "run a chat service" or "run gitlab." This may involve multiple processes, which is fine.

The other reason image authors shy away from process supervisors is they believe a process supervisor must restart failed services, meaning the Docker container will never die.

This does effectively break the Docker ecosystem - most images run one process that will exit when there's an error. By exiting on error, you allow the system administrator to handle failures however they prefer. If your image will never exit, you now need some alternative method of error recovery and failure notification.

Our policy is that if "the thing" fails, then the container should fail, too. We do this by determining which processes can restart, and which should bring down the container. For example, if cron or syslog fails, your container can most likely restart it without any ill effects, but if ejabberd fails, the container should exit so the system administrator can take action.

Our interpretation of "The Docker Way" is thus:

• Containers should do one thing
• Containers should stop when that thing stops

and our init system is designed to do exactly that. Your images will behave like other Docker images and fit in with the existing ecosystem of images.

See "Writing an optional finish script" under the Usage section for details on stopping "the thing."

Init stages

Our overlay init is a properly customized one to run appropriately in containerized environments. This section briefly explains how stages work.

1. stage 1: Its purpose is to set up the image to execute the supervision tree which will handle all the auxiliary services, and to launch stage 2. Stage 1 is where all the black magic happens, all the container setup details that we handle for you so that you don't have to care about them.
2. stage 2: This is where most of the end-user provided files are meant to be executed:
   1. Execute legacy oneshot user scripts contained in /etc/cont-init.d.
   2. Run user s6-rc services declared in the /etc/s6-overlay/user-bundles.d/user bundle, following dependencies. (The services themselves are declared in /etc/s6-overlay/s6-rc.d.)
   3. Copy legacy longrun user services (/etc/services.d) to a temporary directory and have s6 start (and supervise) them.
   4. (if any) Run user s6-rc services declared in the /etc/s6-overlay/user-bundles.d/user2 bundle, following dependencies.
3. stage 3: This is the shutdown stage. When the container is supposed to exit, it will:
   1. Send a TERM signal to all legacy longrun services and, if required, wait for them to exit.
   2. Bring down user s6-rc services in an orderly fashion.
   3. Run any finalization scripts contained in /etc/cont-finish.d.
   4. Send all remaining processes a TERM signal. There should not be any remaining processes anyway.
   5. Sleep for a small grace time, to allow stray processes to exit cleanly.
   6. Send all processes a KILL signal. Then the container exits.

Installation

s6-overlay comes as a set of tarballs that you can extract onto your image. The tarballs you need are a function of the image you use; most people will need the first two, and the other ones are extras you can use at your convenience.

1. s6-overlay-noarch.tar.xz: this tarball contains the scripts implementing the overlay. We call it "noarch" because it is architecture- independent: it only contains scripts and other text files. Everyone who wants to run s6-overlay needs to extract this tarball.
2. s6-overlay-x86_64.tar.xz: replace x86_64 with your system's architecture. This tarball contains all the necessary binaries from the s6 ecosystem, all linked statically and out of the way of your image's binaries. Unless you know for sure that your image already comes with all the packages providing the binaries used in the overlay, you need to extract this tarball.
3. s6-overlay-symlinks-noarch.tar.xz: this tarball contains symlinks to the s6-overlay scripts so they are accessible via /usr/bin. It is normally not needed, all the scripts are accessible via the PATH environment variable, but if you have old user scripts containing shebangs such as #!/usr/bin/with-contenv, installing these symlinks will make them work.
4. s6-overlay-symlinks-arch.tar.xz: this tarball contains symlinks to the binaries from the s6 ecosystem provided by the second tarball, to make them accessible via /usr/bin. It is normally not needed, but if you have old user scripts containing shebangs such as #!/usr/bin/execlineb, installing these symlinks will make them work.
5. syslogd-overlay-noarch.tar.xz: this tarball contains definitions for a syslogd service. If you are running daemons that cannot log to stderr to take advantage of the s6 logging infrastructure, but hardcode the use of the old syslog() mechanism, you can extract this tarball, and your container will run a lightweight emulation of a syslogd daemon, so your syslog logs will be caught and stored to disk.

To install those tarballs, add lines to your Dockerfile that correspond to the functionality you want to install. For instance, most people would use the following:

ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-noarch.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-noarch.tar.xz
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-x86_64.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-x86_64.tar.xz

Make sure to preserve file permissions when extracting (i.e. to use the -p option to tar.)

Usage

The project is distributed as a set of standard .tar.xz files, which you extract at the root of your image. (You need the xz-utils package for tar to understand .tar.xz files; it is available in every distribution, but not always in the default container images, so you may need to apt install xz-utils or apk add xz, or equivalent, before you can expand the archives.)

Afterwards, set your ENTRYPOINT to /init.

Right now, we recommend using Docker's ADD directive instead of running wget or curl in a RUN directive - Docker is able to handle the https URL when you use ADD, whereas your base image might not be able to use https, or might not even have wget or curl installed at all.

From there, you have a couple of options:

• If you want the container to exit when your program exits: run the program as your image's CMD.
• If you want the container to run until told to exit, and your program to be supervised by s6: write a service script for your program.

Using CMD

Using CMD is a convenient way to take advantage of the overlay. Your CMD can be given at build time in the Dockerfile, or at run time on the command line, either way is fine. It will be run as a normal process in the environment set up by s6; when it fails or exits, the container will shut down cleanly and exit. You can run interactive programs in this manner: only the CMD will receive your interactive command, the support processes will be unimpacted.

For example:

FROM busybox
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-noarch.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-noarch.tar.xz
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-x86_64.tar.xz /tmp
RUN tar -C / -Jxpf /tmp/s6-overlay-x86_64.tar.xz
ENTRYPOINT ["/init"]

docker-host $ docker build -t s6demo .
docker-host $ docker run -ti s6demo /bin/sh
/package/admin/s6-overlay/libexec/preinit: notice: /var/run is not a symlink to /run, fixing it
s6-rc: info: service s6rc-oneshot-runner: starting
s6-rc: info: service s6rc-oneshot-runner successfully started
s6-rc: info: service fix-attrs: starting
s6-rc: info: service fix-attrs successfully started
s6-rc: info: service legacy-cont-init: starting
s6-rc: info: service legacy-cont-init successfully started
s6-rc: info: service legacy-services: starting
s6-rc: info: service legacy-services successfully started
/ # ps
PID   USER     TIME  COMMAND
    1 root      0:00 /package/admin/s6/command/s6-svscan -d4 -- /run/service
   17 root      0:00 {rc.init} /bin/sh -e /run/s6/basedir/scripts/rc.init top /bin/sh
   18 root      0:00 s6-supervise s6-linux-init-shutdownd
   20 root      0:00 /package/admin/s6-linux-init/command/s6-linux-init-shutdownd -c /run/s6/basedir -g 3000 -C -B
   24 root      0:00 s6-supervise s6rc-fdholder
   25 root      0:00 s6-supervise s6rc-oneshot-runner
   31 root      0:00 /package/admin/s6/command/s6-ipcserverd -1 -- /package/admin/s6/command/s6-ipcserver-access -v0 -E -l0 -i data/rules -- /packa
   58 root      0:00 /bin/sh
   66 root      0:00 ps
/ # exit
s6-rc: info: service legacy-services: stopping
s6-rc: info: service legacy-services successfully stopped
s6-rc: info: service legacy-cont-init: stopping
s6-rc: info: service legacy-cont-init successfully stopped
s6-rc: info: service fix-attrs: stopping
s6-rc: info: service fix-attrs successfully stopped
s6-rc: info: service s6rc-oneshot-runner: stopping
s6-rc: info: service s6rc-oneshot-runner successfully stopped
docker-host $

Writing a service script

The other way to use a container with s6-overlay is to make your services supervised. You can supervise any number of services; usually they're just support services for the main daemon you run as a CMD, but if that's what you want, nothing prevents you from having an empty CMD and running your main daemon as a supervised service as well. In that case, the daemon will be restarted by s6 whenever it exits; the container will only stop when you tell it to do so, either via a docker stop command, or from inside the container with the /run/s6/basedir/bin/halt command.

There are two ways of making a supervised service. The old way, which is still supported, is to make a "pure s6" service directory. Create a directory with the name of your service in /etc/services.d and put an executable run file into it; this is the file in which you'll put your long-lived process execution. For details of supervision of service directories, and how you can configure how s6 handles your daemon, you can take a look at the servicedir documentation. A simple example would look like this:

/etc/services.d/myapp/run:

#!/command/execlineb -P
nginx -g "daemon off;"

The new way is to make an s6-rc source definition directory in the /etc/s6-overlay/s6-rc.d directory, and add the name of that directory to the user bundle, i.e. create an empty file with the same name in the /etc/s6-overlay/user-bundles.d/user/contents.d directory. The format of a source definition directory is described in this page. Note that you can define longruns, i.e. daemons that will get supervised by s6 just like with the /etc/services.d method, but also oneshots, i.e. programs that will run once and exit. Your main service is probably a longrun, not a oneshot: you probably need a daemon to stick around.

The advantage of this new format is that it allows you to define dependencies between services: if B depends on A, then A will start first, then B will start when A is ready, and when the container is told to exit, B will stop first, then A. If you have a complex architecture where various processes depends on one another, or simply where you have to mix oneshots and longruns in a precise order, this may be for you.

The example above could be rewritten this way:

/etc/s6-overlay/s6-rc.d/myapp/type:

longrun

/etc/s6-overlay/s6-rc.d/myapp/run:

#!/command/execlineb -P
nginx -g "daemon off;"

/etc/s6-overlay/user-bundles.d/user/contents.d/myapp: empty file. (This adds myapp to the set of services that s6-rc will start at container boot.)

/etc/s6-overlay/s6-rc.d/myapp/dependencies.d/base: empty file. (This tells s6-rc to only start myapp when all the base services are ready: it prevents race conditions.)

We encourage you to switch to the new format, but if you don't need its benefits, you can stick with regular service directories in /etc/services.d, it will work just as well.

Setting the exit code of the container to the exit code of your main service

If you run your main service as a CMD, you have nothing to do: when your CMD exits, or when you run docker stop, the container will naturally exit with the same exit code as your service. (Be aware, however, that in the docker stop case, your service will get a SIGTERM, in which case the exit code will entirely depend on how your service handles it - it could trap it and exit 0, trap it and exit something else, or not trap it and let the shell exit its own code for it - normally 130.)

If you run your main service as a supervised service, however, things are different, and you need to tell the container what code to exit with when you send it a docker stop command. To do that, you need to write a finish script:

• If your service is a legacy service in /etc/services.d, you need an executable /etc/services.d/myapp/finish script.
• If your service is an s6-rc one, you need a /etc/s6-overlay/s6-rc.d/myapp/finish file containing your script (the file may or may not be executable).

This finish script will be run when your service exits, and will take two arguments:

• The first argument will be the exit code of your service, or 256 if your service was killed by an uncaught signal.
• The second argument is only meaningful if your service was killed by an uncaught signal, and contains the number of said signal.

In the finish script, you need to write the container exit code you want to the /run/s6-linux-init-container-results/exitcode file - and that's it.

For instance, the finish script for the myapp service above could be something like this:

#!/bin/sh

if test "$1" -eq 256 ; then
  e=$((128 + $2))
else
  e="$1"
fi

echo "$e" > /run/s6-linux-init-container-results/exitcode

When you send a docker stop command to your container, the myapp service will be killed and this script will be run; it will write either myapp's exit code (if myapp catches the TERM signal) or 130 (if myapp does not catch the TERM signal) to the special /run/s6-linux-init-container-results/exitcode file, which will be read by s6-overlay at the end of the container shutdown procedure, and your container will exit with that value.

Fixing ownership and permissions

This section describes a functionality from the versions of s6-overlay that are anterior to v3. fix-attrs is still supported in v3, but is deprecated, for several reasons: one of them is that it's generally not good policy to change ownership dynamically when it can be done statically. Another reason is that it doesn't work with USER containers. Instead of fix-attrs, we now recommend you to take care of ownership and permissions on host mounts offline, before running the container. This should be done in your Dockerfile, when you have all the needed information.

That said, here is what we wrote for previous versions and that is still applicable today (but please stop depending on it):

Sometimes it's interesting to fix ownership & permissions before proceeding because, for example, you have mounted/mapped a host folder inside your container. Our overlay provides a way to tackle this issue using files in /etc/fix-attrs.d. This is the pattern format followed by fix-attrs files:

path recurse account fmode dmode

• path: File or dir path.
• recurse: (Set to true or false) If a folder is found, recurse through all containing files & folders in it.
• account: Target account. It's possible to default to fallback uid:gid if the account isn't found. For example, nobody,32768:32768 would try to use the nobody account first, then fallback to uid 32768 instead. If, for instance, daemon account is UID=2 and GID=2, these are the possible values for account field:
  • daemon: UID=2 GID=2
  • daemon,3:4: UID=2 GID=2
  • 2:2,3:4: UID=2 GID=2
  • daemon:11111,3:4: UID=2 GID=11111
  • 11111:daemon,3:4: UID=11111 GID=2
  • daemon:daemon,3:4: UID=2 GID=2
  • daemon:unexisting,3:4: UID=2 GID=4
  • unexisting:daemon,3:4: UID=3 GID=2
  • 11111:11111,3:4: UID=11111 GID=11111
• fmode: Target file mode. For example, 0644.
• dmode: Target dir/folder mode. For example, 0755.

Here you have some working examples:

/etc/fix-attrs.d/01-mysql-data-dir:

/var/lib/mysql true mysql 0600 0700

/etc/fix-attrs.d/02-mysql-log-dirs:

/var/log/mysql-error-logs true nobody,32768:32768 0644 2700
/var/log/mysql-general-logs true nobody,32768:32768 0644 2700
/var/log/mysql-slow-query-logs true nobody,32768:32768 0644 2700

Executing initialization and finalization tasks

Here is the old way of doing it:

After fixing attributes (through /etc/fix-attrs.d/) and before starting user provided services (through s6-rc or /etc/services.d) our overlay will execute all the scripts found in /etc/cont-init.d, for example:

/etc/cont-init.d/02-confd-onetime:

#!/command/execlineb -P

with-contenv
s6-envuidgid nginx
multisubstitute
{
  import -u -D0 UID
  import -u -D0 GID
  import -u CONFD_PREFIX
  define CONFD_CHECK_CMD "/usr/sbin/nginx -t -c {{ .src }}"
}
confd --onetime --prefix="${CONFD_PREFIX}" --tmpl-uid="${UID}" --tmpl-gid="${GID}" --tmpl-src="/etc/nginx/nginx.conf.tmpl" --tmpl-dest="/etc/nginx/nginx.conf" --tmpl-check-cmd="${CONFD_CHECK_CMD}" etcd

This way is still supported. However, there is now a more generic and efficient way to do it: writing your oneshot initialization and finalization tasks as s6-rc services, by adding service definition directories in /etc/s6-overlay/s6-rc.d, making them part of the user bundle (so they are actually started when the container boots), and making them depend on the base bundle (so they are only started after base).

All the information on s6-rc can be found here.

When the container is started, the operations are performed in this order:

• (deprecated) Attribute fixing is performed according to files in /etc/fix-attrs.d.
• (legacy) One-shot initialization scripts in /etc/cont-init.d are run sequentially.
• Services in the user bundle are started by s6-rc, in an order defined by dependencies. Services can be oneshots (initialization tasks) or longruns (daemons that will run throughout the container's lifetime). If the services depend on base, they are guaranteed to start at this point and not earlier; if they do not, they might have been started earlier, which may cause race conditions - so it's recommended to always make them depend on base.
• (legacy) Longrun services in /etc/services.d are started.
• Services in the user2 bundle with the correct dependency are started. (If you're not using /etc/services.d, or if you're not sure, you don't need to care about user2 at all - just stick to user.)

When the container is stopped, either because the admin sent a stop command or because the CMD exited, the operations are performed in the reverse order:

• Services in the user2 bundle with the correct dependency are stopped.
• (legacy) Longrun services in /etc/services.d are stopped.
• All s6-rc services are stopped, in an order defined by dependencies. For oneshots, that means that the down script in the source definition directory is executed; that's how s6-rc can perform finalization tasks.
• (legacy) One shot finalization scripts in /etc/cont-finish.d are run sequentially.

The point of the user2 bundle is to allow user services declared in it to start after the /etc/services.d ones; but in order to do so, every service in user2 needs to declare a dependency to legacy-services. In other words, for a service foobar to start late, you need to:

• Define it in /etc/s6-overlay/s6-rc.d/foobar like any other s6-rc service.
• Add an /etc/s6-overlay/s6-rc.d/foobar/dependencies.d/legacy-services file
• Add an /etc/s6-overlay/user-bundles.d/user2/contents.d/foobar file.

That will ensure that foobar will start after everything in /etc/services.d.

Writing an optional finish script

By default, services created in /etc/services.d will automatically restart. If a service should bring the container down, you should probably run it as a CMD instead; but if you'd rather run it as a supervised service, then you'll need to write a finish script, which will be run when the service is down; to make the container stop, the /run/s6/basedir/bin/halt command must be invoked. Here's an example finish script:

/etc/services.d/myapp/finish:

#!/command/execlineb -S0

foreground { redirfd -w 1 /run/s6-linux-init-container-results/exitcode echo 0 }
/run/s6/basedir/bin/halt

The first line of the script writes 0 to the /run/s6-linux-init-container-results/exitcode file. The second line stops the container. When you stop the container via the /run/s6/basedir/bin/halt command run from inside the container, /run/s6-linux-init-container-results/exitcode is read and its contents are used as the exit code for the docker run command that launched the container. If the file doesn't exist, or if the container is stopped with docker stop or another reason, that exit code defaults to 0.

It is possible to do more advanced operations in a finish script. For example, here's a script from that only brings down the service when it exits nonzero:

/etc/services.d/myapp/finish:

#!/command/execlineb -S1
if { eltest ${1} -ne 0 -a ${1} -ne 256 }
/run/s6/basedir/bin/halt

Note that in general, finish scripts should only be used for local cleanups after a daemon dies. If a service is so important that the container needs to stop when it dies, we really recommend running it as the CMD.

Logging

Every service can have its dedicated logger. A logger is a s6 service that automatically reads from the stdout of your service, and logs the data to an automatically rotated file in the place you want. Note that daemons usually log to stderr, not stdout, so you should probably start your service's run script with exec 2>&1 in shell, or with fdmove -c 2 1 in execline, in order to catch stderr.

s6-overlay provides a utility called logutil-service which is a wrapper over the s6-log program. This helper does the following:

• read how s6-log should proceed reading the logging script contained in S6_LOGGING_SCRIPT
• drop privileges to the nobody user (defaulting to 65534:65534 if it doesn't exist)
• clean all the environments variables
• execute into s6-log.

s6-log will then run forever, reading data from your service and writing it to the directory you specified to logutil-service.

Please note:

• Since the privileges are dropped automatically, there is no need to switch users with s6-setuidgid
• You should ensure the log folder either:
  • exists, and is writable by the nobody user
  • does not exist, but the parent folder is writable by the nobody user.

You can create log folders in cont-init.d scripts, or as s6-rc oneshots. Here is an example of a logged service myapp implemented the old way:

/etc/cont-init.d/myapp-log-prepare:

#!/bin/sh -e
mkdir -p /var/log/myapp
chown nobody:nogroup /var/log/myapp
chmod 02755 /var/log/myapp

/etc/services.d/myapp/run:

#!/bin/sh
exec 2>&1
exec mydaemon-in-the-foreground-and-logging-to-stderr

/etc/services.d/myapp/log/run:

#!/bin/sh
exec logutil-service /var/log/myapp

And here is the same service, myapp, implemented in s6-rc.

/etc/s6-overlay/s6-rc.d/myapp-log-prepare/dependencies.d/base: empty file

/etc/s6-overlay/s6-rc.d/myapp-log-prepare/type:

oneshot

/etc/s6-overlay/s6-rc.d/myapp-log-prepare/up:

if { mkdir -p /var/log/myapp }
if { chown nobody:nogroup /var/log/myapp }
chmod 02755 /var/log/myapp

/etc/s6-overlay/scripts/myapp-log-prepare

#!/bin/sh -e
mkdir -p /var/log/myapp
chown nobody:nogroup /var/log/myapp
chmod 02755 /var/log/myapp

/etc/s6-overlay/s6-rc.d/myapp/dependencies.d/base: empty file

/etc/s6-overlay/s6-rc.d/myapp-log/dependencies.d/myapp-log-prepare: empty file

/etc/s6-overlay/s6-rc.d/myapp/type:

longrun

/etc/s6-overlay/s6-rc.d/myapp/run:

#!/bin/sh
exec 2>&1
exec mydaemon-in-the-foreground-and-logging-to-stderr

/etc/s6-overlay/s6-rc.d/myapp/producer-for:

myapp-log

/etc/s6-overlay/s6-rc.d/myapp-log/type:

longrun

/etc/s6-overlay/s6-rc.d/myapp-log/run:

#!/bin/sh
exec logutil-service /var/log/myapp

/etc/s6-overlay/s6-rc.d/myapp-log/consumer-for:

myapp

/etc/s6-overlay/s6-rc.d/myapp-log/pipeline-name:

myapp-pipeline

/etc/s6-overlay/user-bundles.d/user/contents.d/myapp-pipeline: empty file

That's a lot of files! A summary of what it all means is:

• myapp-log-prepare is a oneshot, preparing the logging directory. It is a dependency of myapp-log, so it will be started before myapp-log.
• myapp is a producer for myapp-log and myapp-log is a consumer for myapp, so what myapp writes to its stdout will go to myapp-log's stdin. Both are longruns, i.e. daemons that will be supervised by s6.
• The myapp | myapp-log pipeline is given a name, myapp-pipeline, and this name is declared as a part of the user bundle, so it will be started when the container starts.
• myapp-log-prepare, myapp-log and myapp all depend on the base bundle, which means they will only be started when the system is actually ready to start them.

It really accomplishes the same things as the /etc/cont-init.d plus /etc/services.d method, but it's a lot cleaner underneath, and can handle much more complex dependency graphs, so whenever you get the opportunity, we recommend you familiarize yourself with the s6-rc way of declaring your services and your loggers. The full syntax of a service definition directory, including declaring whether your service is a longrun or a oneshot, declaring pipelines, adding service-specific timeouts if you need them, etc., can be found here.

Dropping privileges

When it comes to executing a service, no matter whether it's a service or a logger, a good practice is to drop privileges before executing it. s6 already includes utilities to do exactly these kind of things:

In execline:

#!/command/execlineb -P
s6-setuidgid daemon
myservice

In sh:

#!/bin/sh
exec s6-setuidgid daemon myservice

If you want to know more about these utilities, please take a look at: s6-setuidgid, s6-envuidgid, and s6-applyuidgid.

Container environment

If you want your custom script to have container environments available: you can use the with-contenv helper, which will push all of those into your execution environment, for example:

/etc/cont-init.d/01-contenv-example:

#!/command/with-contenv sh
env

This script will output the contents of your container environment.

Read-Only Root Filesystem

Recent versions of Docker allow running containers with a read-only root filesystem. If your container is in such a case, you should set S6_READ_ONLY_ROOT=1 to inform s6-overlay that it should not attempt to write to certain areas - instead, it will perform copies into a tmpfs mounted on /run.

Note that s6-overlay assumes that:

• /run exists and is writable. If it is not, it will attempt to mount a tmpfs there.
• /var/run is a symbolic link to /run, for compatibility with previous versions. If it is not, it will make it so.

In general your default docker settings should already provide a suitable tmpfs in /run.

Customizing s6-overlay behaviour

It is possible somehow to tweak s6-overlay's behaviour by providing an already predefined set of environment variables to the execution context:

• PATH (default = /command:/usr/bin:/bin): this is the default PATH that all the services in the container, including the CMD, will have. Set this variable if you have a lot of services that depend on binaries stored in another directory, e.g. /usr/sbin. Note that /command, /usr/bin and /bin will always be added to that path if they're not already in the one you provide.
• S6_KEEP_ENV (default = 0): if set, then environment is not reset and whole supervision tree sees original set of env vars. It switches with-contenv into a nop.
• S6_LOGGING (default = 0):
  • 0: Outputs everything to stdout/stderr.
  • 1: Uses an internal catch-all logger and persists everything on it, it is located in /var/log/s6-uncaught-logs. Anything run as a CMD is still output to stdout/stderr.
  • 2: Uses an internal catch-all logger and persists everything on it, including the output of CMD. Absolutely nothing is written to stdout/stderr.
• S6_CATCHALL_USER (default = root): if set, and if S6_LOGGING is 1 or 2, then the catch-all logger is run as this user, which must be defined in your image's /etc/passwd. Every bit of privilege separation helps a little with security.
• S6_BEHAVIOUR_IF_STAGE2_FAILS (default = 0): determines what the container should do if one of the service scripts fails. This includes:
  • if the early stage2 hook exits nonzero (by default there's no hook)
  • if anything fails in fix-attrs
  • if any old-style /etc/cont-init.d or new-style s6-rc oneshot fails
  • if any old-style /etc/services.d or new-style s6-rc longrun is marked as expecting readiness notification, and fails to become ready in the allotted time (see S6_CMD_WAIT_FOR_SERVICES_MAXTIME below). The valid values for S6_BEHAVIOUR_IF_STAGE2_FAILS are the following:
  • 0: Continue silently even if a script has failed.
  • 1: Continue but warn with an annoying error message.
  • 2: Stop the container.
• S6_KILL_FINISH_MAXTIME (default = 5000): How long (in milliseconds) the system should wait, at shutdown time, for a script in /etc/cont-finish.d to finish naturally. After this duration, the script will be sent a SIGKILL. Bear in mind that scripts in /etc/cont.finish.d are run sequentially, and the shutdown sequence will potentially wait for S6_KILL_FINISH_MAXTIME milliseconds for each script.
• S6_SERVICES_READYTIME (default = 50): With services declared in /etc/services.d, there is an unavoidable race condition between the moment when services are started and the moment when they can be tested for readiness. To avoid that race, we sleep a little time, by default 50 milliseconds, before testing for readiness. If your machine is slow or very busy, you may get errors looking like s6-svwait: fatal: unable to s6_svstatus_read: No such file or directory. In that case, you should increase the sleeping time, by declaring it (in milliseconds) in the S6_SERVICES_READYTIME variable. Note that it only concerns /etc/services.d; s6-rc is immune to the race condition.
• S6_SERVICES_GRACETIME (default = 3000): How long (in milliseconds) s6 should wait, at shutdown time, for services declared in /etc/services.d to die before proceeding with the rest of the shutdown.
• S6_KILL_GRACETIME (default = 3000): How long (in milliseconds) s6 should wait, at the end of the shutdown procedure when all the processes have received a TERM signal, for them to die before sending a KILL signal to make sure they're dead.
• S6_LOGGING_SCRIPT (default = "n20 s1000000 T"): This env decides what to log and how, by default every line will prepend with ISO8601, rotated when the current logging file reaches 1mb and archived, at most, with 20 files.
• S6_CMD_ARG0 (default = not set): Value of this env var will be prepended to any CMD args passed by docker. Use it if you are migrating an existing image to s6-overlay and want to make it a drop-in replacement: setting this variable to the value of a previously used ENTRYPOINT will help you transition.
• S6_CMD_USE_TERMINAL (default = 0): Set this value to 1 if you have a CMD that needs a terminal for its output (typically when you're running your container with docker run -it), and you have set S6_LOGGING to a nonzero value. This setting will make your CMD actually output to your terminal; the drawback is that its output will not be logged. By default (when this variable is 0 or not set), the stdout and stderr of your CMD are logged when S6_LOGGING is nonzero, which means they go to a pipe even if you're running it in an interactive terminal.
• S6_FIX_ATTRS_HIDDEN (default = 0): Controls how fix-attrs.d scripts process files and directories.
  • 0: Hidden files and directories are excluded.
  • 1: All files and directories are processed.
• S6_CMD_WAIT_FOR_SERVICES (default = 0): By default when the container starts, services in /etc/services.d will be started and execution will proceed to starting the user2 bundle and the CMD, if any of these is defined. If S6_CMD_WAIT_FOR_SERVICES is nonzero, however, the container starting sequence will wait until the services in /etc/services.d are ready before proceeding with the rest of the sequence. Note that this is only significant if the services in /etc/services.d notify their readiness to s6.
• S6_CMD_WAIT_FOR_SERVICES_MAXTIME (default = 0, i.e. infinite): The maximum time (in milliseconds) the services could take to bring up before proceding to CMD executing. Set this variable to a positive value if you have services that can potentially block indefinitely and you prefer the container to fail if not everything is up after a given time. Note that this value also includes the time setting up legacy container initialization (/etc/cont-init.d) and services (/etc/services.d), so take that into account when computing a suitable value. In versions of s6-overlay up to 3.1.6.2, the default was 5000 (five seconds), but it caused more unwanted container failures than it solved issues, so now there's no timeout by default: s6-overlay will wait as long as is necessary for all the services to be brought up.
• S6_READ_ONLY_ROOT (default = 0): When running in a container whose root filesystem is read-only, set this env to 1 to inform init stage 2 that it should copy user-provided initialization scripts from /etc to /run/s6/etc before it attempts to change permissions, etc. See Read-Only Root Filesystem for more information.
• S6_SYNC_DISKS (default = 0): Set this env to 1 to inform init stage 3 that it should attempt to sync filesystems before stopping the container. Note: this will likely sync all filesystems on the host.
• S6_STAGE2_HOOK (default = none): If this variable exists, its contents will be interpreted as a shell excerpt that will be run in the early stage 2, before services are started. This can be used, for instance, to dynamically patch the service database at run-time right before it is compiled and run. If the hook program exits nonzero and S6_BEHAVIOUR_IF_STAGE2_FAILS is 2 or more, the container will stop instantly. Please note that running the wrong hook program may prevent your container from starting properly, or may endanger your security; so only use this if you know exactly what you are doing. When in doubt, leave this variable undefined.
• S6_VERBOSITY (default = 2): controls the verbosity of s6-rc, and potentially other tools, at container start and stop time. The default, 2, is normally verbose: it will list the service start and stop operations. You can make the container quieter by decreasing this number: 1 will only print warnings and errors, and 0 will only print errors. You can also make the container more verbose, i.e. print tracing and debug information, by increasing this number up to 5, but the output will quickly become very noisy, and most people shouldn't need this.
• S6_CMD_RECEIVE_SIGNALS (default = 0): decides whether signals sent to the container should be sent to the container's pid 1 or to the CMD. By default, when you perform for instance a docker stop, a TERM signal will be sent to the container's pid 1, which will trigger the full container shutdown sequence - but if a CMD is present, it will be among the last processes to be killed, only when everything else is down and the container is about to exit. If this variable is 1 or more, signals are diverted from pid 1 to the CMD, which means that docker stop will send a SIGTERM to the CMD instead, and the container will only trigger its shutdown procedure when the CMD is dead. Note that only SIGTERM, SIGQUIT, SIGINT, SIGUSR1, SIGUSR2, SIGPWR and SIGWINCH are diverted; other signals either are ignored or cannot be diverted and are necessarily handled by pid 1. Please be aware that using this option may prevent interactive CMDs from working at all - in other words, if you're running an interactive CMD in a terminal, don't set this variable; but that should be fine since in this case you already have interactive ways of stopping your CMD.
• S6_YES_I_WANT_A_WORLD_WRITABLE_RUN_BECAUSE_KUBERNETES (default = 0): yes, it's a mouthful, and yes, we're deadly serious. If you set this variable to 1 (or any non-zero value), s6-overlay will accept to boot in a situation where /run belongs to uid 0 and is world-writable (permissions 0777) but the container is run as a non-root user. This is the configuration enforced by some Kubernetes environments, and it is completely insecure, except that Kubernetes ensures isolation by its own security mechanisms; so, if you're in such a situation, set that variable, and s6-overlay will still print a (literally) bright red warning, but it will boot. Otherwise, by default, s6-overlay refuses to run when it encounters such a insecure setup that it doesn't have the necessary privileges to fix.
• S6_WAIT_FOR_CLOSING_FD (default = 0): if this variable is set to 3 or more, then this number represents a file descriptor. Your container manager must spawn your container with this file descriptor open, and use it as a synchronization mechanism, as in, when reading this file descriptor returns EOF, it means that the container manager is ready and the container can proceed. Some Docker setups use fd 3 for this; some systemd setups use 4. Read your container manager's documentation to check if such synchronization is necessary; when the variable is set, s6-overlay will follow that protocol. If the variable is unset, or set to 0 (or 1 or 2 but don't do this), s6-overlay will not attempt to synchronize with the container manager, it will just boot your container as fast as possible.
• S6_RUNTIME_BUNDLEDIR (default=/etc/s6-overlay/user-bundles.d): if this variable is set, then it must contain a path to a directory containing user and user2 subdirectories, themselves being proper bundle definitions similar to /etc/s6-overlay/user-bundles.d/user. At container initialization time, the definitions for the user and user2 bundles will be read from $S6_RUNTIME_BUNDLEDIR/user and $S6_RUNTIME_BUNDLEDIR/user2.
• S6_RUNTIME_PROFILE: if this variable is set, then it must contain the name of a subdirectory of /etc/cont-profile.d, which contains a full replacement for /etc/s6-overlay. In other words, if S6_RUNTIME_PROFILE is foobar, then /etc/cont-profile.d/foobar/s6-overlay must exist and contain s6-rc.d and user-bundles.d directories, defining a full replacement for all the services in /etc/s6-overlay. This is an older, more powerful, more cumbersome customization switch than S6_RUNTIME_BUNDLEDIR; it is likely that you don't need it and that S6_RUNTIME_BUNDLEDIR fits your needs.

syslog

If software running in your container requires syslog, extract the syslogd-overlay-noarch.tar.xz tarball: that will give you a small syslogd emulation. Logs will be found under various subdirectories of /var/log/syslogd, for instance messages will be found in the /var/log/syslogd/messages/ directory, the latest logs being available in the /var/log/syslogd/messages/current file. Logging directories are used rather than files so that logs can be automatically rotated without race conditions (that is a feature of s6-log).

It is recommended to add syslog and sysllog users to your image, for privilege separation; the syslogd emulation processes will run as these users if they exist. Otherwise they will default to 32760:32760 and 32761:32761, numeric uids/gids that may already exist on your system.

Performance

• The noarch and symlinks tarballs are all tiny. The biggest tarball is the one that contains the binaries; it's around 650 kB.
• Uncompressed on a tmpfs, the overlay scripts use about 120 kB, and the binaries for x86_64 use about 5.7 MB.
• We haven't yet measured the time it takes for the container to be up and running once you run docker run, but you will notice it's fast. Faster than previous versions of s6-overlay, with fewer delays. And if you convert your /etc/cont-init.d scripts to the s6-rc format, they will be able to run in parallel, so you will gain even more performance. If you have benchmarks, please send them to us!

Verifying Downloads

The s6-overlay releases have a checksum files you can use to verify the download using SHA256:

ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-noarch.tar.xz /tmp
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-noarch.tar.xz.sha256 /tmp
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-x86_64.tar.xz /tmp
ADD https://github.com/just-containers/s6-overlay/releases/download/v${S6_OVERLAY_VERSION}/s6-overlay-x86_64.tar.xz.sha256 /tmp
RUN cd /tmp && sha256sum -c *.sha256

Notes

USER directive

As of version 3.2.3.2, s6-overlay has limited support for running as a user other than root:

• Tools like fix-attrs and logutil-service are unlikely to work (they rely on being able to change UIDs).
• The syslogd emulation will not work.

Generally speaking, if you're running a simple container with a main application and one or two support services, you may benefit from the USER directive if that is your preferred way of running containers. However, if you're running more than a few services, or daemons that expect a real system with complete Unix infrastructure, then USER is probably not a good idea and you would benefit more from using privilege separation between services in your container.

Terminal support

Generally speaking, you should not run your containers with docker run -it. It is bad practice to have console access to your containers. That said, if your CMD is interactive and needs a terminal, s6-overlay will try to support it whenever possible, but the nature of terminals makes it difficult to ensure that everything works perfectly in all cases.

In particular, if you are stacking virtualization environments and other layers already have their own kludges for terminals - for instance, if you are running s6-overlay under qemu - then it is almost guaranteed that docker run -it will not work. However, once the container is running, you should always be able to access an interactive shell inside it via docker exec -it containername /bin/sh.

The same caveats apply to stopping containers with ^C. Normally containers are stopped via docker stop, or when the CMD exits; ^C is not an officially supported method of stopping them. s6-overlay tries to exit cleanly on ^C, whether the container is running with -it or not, but there will be cases where it is unfortunately impossible.

Releases

Over on the releases tab, we have a number of tarballs:

• s6-overlay-noarch.tar.xz: the s6-overlay scripts.
• s6-overlay-${arch}.tar.xz: the binaries for platform ${arch}. They are statically compiled and will work with any Linux distribution.
• s6-overlay-symlinks-noarch.tar.xz: /usr/bin symlinks to the s6-overlay scripts. Totally optional.
• s6-overlay-symlinks-arch.tar.xz: /usr/bin symlinks to the skaware binaries. Totally optional.
• syslogd-overlay-noarch.tar.xz: the syslogd emulation. Totally optional.
• s6-overlay-${version}.tar.xz: the s6-overlay source. Download it if you want to build s6-overlay yourself.

We have binaries for at least x86_64, aarch64, arm32, i486, i686, riscv64, and s390x. The full list of supported arches can be found in conf/toolchains.

Which architecture to use depending on your TARGETARCH

The ${arch} part in the s6-overlay-${arch}.tar.xz tarball uses the naming conventions of gcc, which are not the ones that Docker uses. (Everyone does something different in this field depending on their needs, and no solution is better than any other, but the Docker one is worse than others because its naming is inconsistent. The gcc convention is better for us because it simplifies our builds greatly and makes them more maintainable.)

The following table should help you find the right tarball for you if you're using the TARGETARCH value provided by Docker:

If you need another architecture, ask us and we'll try to make a toolchain for it. In particular, we know that armv7 is a mess and needs a flurry of options depending on your precise target (and this is one of the reasons why the Docker naming system isn't good, although arguably the gcc naming system isn't much better on that aspect).

Contributing

Any way you want! Open issues, open PRs, we welcome all contributors!

Building the overlay yourself

• Download the [s6-overlay source].
• Check the conf/defaults.mk file for variables you may want to change. Do not modify the file itself.
• Call make followed by your variable assignments. Example: make ARCH=riscv64-linux-musl to build the overlay for RISCV64.
• The tarballs will be in the output subdirectory, unless you changed the OUTPUT variable.

Upgrade Notes

Please see CHANGELOG.