Difference between revisions of "Runit"

Revision as of 01:39, 7 December 2011

Runit is a process supervisor. It includes runit-init, which can replace sysv's init as pid1, or can be run from inittab or your init system of choice. Runit's simple collection of tools can be used to build flexible dependency structures and distributed systems, or blazing fast parallel runlevel changes (including the initial boot).

See G. Pape's Runit Page for a complete description, but follow the installation instructions below for your Arch system.

Installation

To replace init with runit-init

• install runit-dietlibc and runit-run from the AUR
• choose/create a default runlevel (see Run Levels)
• reboot

If you just want to get your feet wet and not replace init just yet, runit-dietlibc can be installed side-by-side with the regular Arch initscripts, providing just process supervision of those services you put in /var/service.

• install runit-dietlibc from the AUR
• start runsvdir /var/service using your current init scheme (inittab/rc.local/systemd, whatever)

You may install runit-services and/or runit-scripts to install many services built for runit's style of supervision.

The runit-services package puts services in /etc/sv and uses /usr/bin/rsvlog as a logger (it's a shell script, take a look and modify to taste, improvements welcome).

runit-scripts puts many new runlevels and symlinks them to the service directories it creates in /etc/runit/runsvdir/all, and uses its own /usr/bin/nsvlog script for logging.

Using runit

The Tools

• sv - used for controlling services, getting status of services, and dependency checking.
• chpst- control of a process environment, including memory caps, limits on cores, data segments, environments, user/group privileges, and more.
• runsv - supervises a process, and optionally a log service for that process.
• svlogd - a simple but powerful logger, includes auto-rotation based on different methods (time, size, etc), post-processing, pattern matching, and socket (remote logging) options. Say goodbye to logrotate and the need to stop your services to rotate logs.
• runsvchdir - changes service levels (runlevels, see below).
• runsvdir - starts a supervision tree
• runit-init - PID 1, tiny, does almost nothing, dietlibc staticly compiled. Just what you want your PID 1 to be.

See the manpages for usage details not covered below.

The Extras

Added by runit-dietlibc and runit-run

• /etc/runit/1 - bootstraps the system using arch rc scripts
• /etc/runit/2 - starts single or multi-user runlevels using arch's rc.single or rc.multi
• /etc/runit/3 - brings the system down using arch's rc scripts
• /etc/runit/runsvdir/* - various runlevels
• /usr/bin/rsvlog - a wrapper to svlogd meant to be symlinked as 'run' in a log service
• /etc/sv/* - the service directories available (more available here when you install runit-services-git)

Added by runit-scripts

• /etc/runit/1_new - meant to be an alternate way to bootstrap, does not necessarily use arch boot scripts
• /etc/runit/2_new - single/multi user runlevels (not based on arch scripts)
• /etc/runit/3_new - take the system down
• /etc/runit/runsvdir/all - every service directory available
• /etc/runit/runsvdir/* - various runlevels
• /usr/bin/nsvlog - wrapper meant to be symlinked as 'run' in a log service

Run Levels and Service Directories

Runit uses directories of symlinks to specify runlevels, other than the 3 main ones, which are defined in /etc/runit/1, 2, and 3.

1 bootstraps the system, 2 starts runsvdir on /service, and 3 stops the system.

While in run level 2, you are not constrained to any amount of service levels (equivalent to runlevels in sysvinit). You can runschdir to any directory (full of service directory symlinks) you've made in /etc/runit/runsvdir/. This becomes very handy in cases where you have an HA (Failover) setup, and you have one machine that can take over services for many other machines, simply by runsvchdir <theservicedir>.

You can also run trees of dependent service levels by having user-level supervision directories. See User Level Services below.

By default, the runit-run-git packages uses a very minimal service set, defined in /etc/runit/runsvdir/runit-default and symlinked to /etc/runit/runsvdir/default.

It only gives gettys on tty2 and tty3, so you will boot to just console scroll and a tidy 'runsvchdir: default: current'. This means when you start X it will be on tty4.

To go back to the standard arch consoles, use the arch-default directory in /etc/runit/runsvdir (symlink it to /etc/runit/runsvdir/default). Better yet, create your own directory in /etc/runit/runsvdir and add the symlinks you want for just the services you desire. Remember to take any services you start with runit out of DAEMONS in /etc/rc.conf, they do not need to be started there, and runit will allow parallel startup without backgrounding them.

General Use

For convenience I'll be using /service as the service directory in these examples. Since this has not been accepted by FHS, it is only made available as a symlink in the runit-run-git package. This allows importing of /service scripts written by others without as much fuss. If you only install runit-dietlibc, you would use /var/service as your service directory, or make the /service symlink to /var/service yourself.

Listing running services

Template:Cli

   run: /service/agetty-2: (pid 4120) 7998s
   run: /service/agetty-3: (pid 4119) 7998s
   run: /service/bougyman: (pid 4465) 7972s
   run: /service/bougyx: (pid 4135) 7998s; run: log: (pid 4127) 7998s
   run: /service/cron: (pid 4137) 7998s; run: log: (pid 4122) 7998s
   run: /service/dialer: (pid 4121) 7998s
   run: /service/qmail: (pid 4138) 7998s; run: log: (pid 4126) 7998s
   run: /service/smtpd: (pid 4136) 7998s; run: log: (pid 4125) 7998s
   run: /service/socklog-klog: (pid 4139) 7998s; run: log: (pid 4132) 7998s
   run: /service/socklog-unix: (pid 4133) 7998s; run: log: (pid 4124) 7998s
   run: /service/ssh: (pid 4134) 7998s; run: log: (pid 4123) 7998s

Services should live in /etc/sv; however, the runit-scripts package puts a bunch of common ones in /etc/runit/runsvdir/all. We're working to replace/convert these to /etc/sv in the runit-services-git package.

Create and Start a service:

Template:Cli

Stops a service immediately (would still start on next boot):

Template:Cli

Restarts a service:

Template:Cli

Reloads a service:

Template:Cli

Shows status of a service and it's log service:

Template:Cli

Stops a service, and disables it (won't start next boot):

Template:Cli

Refer to man sv for more details.

Shut down the system

Template:Cli

Reboot the system

Template:Cli

We've symlinked /sbin/runit-init to /sbin/init in runit-run, so as not to have to change grub stuff. The original sysvinit can be started (even after a runit-run install) with init=/sbin/init.sysv appended to your kernel command line.

User Level Services

You can extend the supervision tree by starting a runsvdir as a specific user, giving that user control of their own supervise tree.

Add a user level service tree

Template:Cli

Create /etc/sv/homes/joeuser/run with the following:

   #!/bin/sh
   export PATH=/home/joeuser/bin:$PATH # optional, if your services rely on binaries in ~/bin
   exec 2>&1 \
     sudo -H -u joeuser runsvdir -P /home/joeuser/service 'log:...................................................................................................................................' # Requires sudo, of course
     

Template:Cli

Then symlink /etc/sv/homes/joeuser to /service and any service joe puts in ~/service will start, as him, with his environment.

(the .......... represent placeholders, the proceess will print stdout/err every 5 seconds for each placeholder . you use in this case)

Create an X session service for a user

Template:Cli

Create the /etc/sv/joeuserX/run script with the following

   #!/bin/sh
   exec 2>&1 \
     su -c xinit - joeuser

Template:Cli

Then symlink /etc/sv/joeuserX to /service. joe's X session will now always run (in this runlevel). To protect it using joe's ssh passphrase, use the following in your .xinitrc:

   #!/bin/sh
   ...
   SNIP
   ...
   xscreensaver&
   eval $(keychain --eval)
   exec sh -c \
     'SSH_ASKPASS=/usr/lib/openssh/ssh-askpass-fullscreen ssh-add < /dev/null \
      && exec stumpwm'

Replace the 'stumpwm' with the command to launch your window manager or desktop environment.

Requires the 'keychain' and 'ssh-askpass-fullscreen' packages, or you could replace 'eval $(keychain)' with 'eval $(ssh-agent) and replace ssh-askpass-fullscreen with any ssh passphrase asker. The fullscreen version guarantees protection of your desktop, so we prefer that. This also exports your key to all your x apps, so you do not need another keychain manager for ssh. In addition, 'keychain' (as opposed to just ssh-agent) supports gpg passphrase caching, as well, not just the ssh keys.

Advanced Recipes

Running a read-only Postgresql Slave database in-memory

This recipe was created for a small but vital database which required very high read throughput. To sort it out we use Postgresql's Streaming Replication and Hot Standby mode.

Requirements

• Postgresql 9.0 or above
• runit-services (includes /etc/sv/postgresql)
• Rsync (for initial replication)

Instructions

1. Create /etc/sv/pg_mem/log directory {bc|# mkdir -p /etc/sv/pg_mem/log}

2. Create three new files

/etc/sv/pg_shm:

   #!/bin/sh -e
   sleep 3 # Give postgresql a chance to start and replay any transactions
   
   . /etc/conf.d/pg_shm # Read any conf vars
   PG_DISK_ROOT=/var/lib/postgres # Where the 'master' data directory lives
   
   [ -d "$PGROOT" ] || mkdir -p "$PGROOT" # Create the new $PGROOT if it does not exist
   
   sv -w7 c postgresql 2>&1
   
   # Stop the main postgres from making changes by enttering backup mode
   psql -U postgres -c "SELECT pg_start_backup('seed',true)" 2>&1
   # Sync the main postgres data dir to our new $PGROOT
   rsync --progress --delete -a "$PG_DISK_ROOT/data" "$PGROOT/" --exclude=postmaster.pid 2>&1
   # Allow changes on the primary server again
   psql -U postgres -c "SELECT pg_stop_backup()" 2>&1
   
   # Set up the hot standby mode on the slave server
   echo "hot_standby = 'on'" >> "$PGROOT/data/postgresql.conf"
   echo "port = $PGPORT" >> "$PGROOT/data/postgresql.conf"
   echo "standby_mode = 'on'" >> "$PGROOT/data/recovery.conf"
   echo "primary_conninfo = 'host=localhost port=5432 user=postgres'" >> "$PGROOT/data/recovery.conf"
   echo "trigger_file = '/tmp/stop_replication'" >> "$PGROOT/data/recovery.conf"
   echo "restore_command = 'cp /var/lib/postgres/archive/%f \"%p\"'" >> "$PGROOT/data/recovery.conf"
   
   exec chpst -u postgres /usr/bin/postgres -D "$PGROOT/data" -c config_file="$PGROOT/data/postgresql.conf" 2>&1

Which requires /etc/conf.d/pg_shm:

PGROOT=/dev/shm/pg_mem
PGPORT=5434
PGLOG="/var/log/pg_mem.log"

as well as a file in /etc/sv/postgresql (or wherever your postgresql service directory lives) named 'finish':

#/bin/sh
#sv -v i pg_shm

3. Edit /var/lib/postgres/data/postmaster.conf, to enable wal archiving. See this The PostgreSQL page, steps 3 and 4, for detailed instructions on this.

4. Restart postgresql {bc|# sv i postgresql}

5. Start pg_shm (replace /service with your service directory, if it differs) {bc|# ln -s /etc/sv/pg_shm /service}

That's it, you'll have a replica of your postgresql on-disk database published on port 5434, in read-only mode from the memory space utilized from /dev/shm.