www.nico.schottelius.org/software/cinit/browse_source/cinit-0.3pre18/doc/user/current-init-problems.text
Nico Schottelius ca81f97577 the key itself plus the cinit release
Signed-off-by: Nico Schottelius <nico@ikn.schottelius.org>
2009-11-07 21:11:05 +01:00

170 lines
4.6 KiB
Text

Currently available init systems and their problems
===================================================
Nico Schottelius <nico-cinit__@__schottelius.org>
0.1, for cinit 0.3, Initial version from 2006-12-11
:Author Initials: NS
This document describes some other init systems, especially
the "traditional" ones and describes their problems.
Introduction
-------------
some history, sysv, bsd,
minit, runit, initng, cinit, upstart
Implementations
---------------
List of current implementations as of 2006-12-12.
Linux Sys-V-Init
~~~~~~~~~~~~~~~~
Debian Sid
/etc/inittab
"First script" /etc/init.d/rcS (specified
/etc/init.d/rcS: Bourne shell script text executable
Runlevels: 0-6, different states
--------------------------------------------------------------------------------
[9:51] hydrogenium:cinit-0.3pre6% /sbin/runlevel
N 2
--------------------------------------------------------------------------------
Depending on runlevel, the init calls /etc/init.d/rc $runlevel.
This is manually configured in /etc/inittab
/etc/init.d/rc: Bourne shell script text executable
/etc/init.d/rc
Calls symlinks in /etc/rc?.d. Each runlevel a directory:
--------------------------------------------------------------------------------
[9:51] hydrogenium:cinit-0.3pre6% ls -ld /etc/rc*.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc0.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc1.d
drwxr-xr-x 3 root root 4096 2006-12-01 16:56 /etc/rc2.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc3.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc4.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc5.d
drwxr-xr-x 2 root root 4096 2006-12-01 16:56 /etc/rc6.d
drwxr-xr-x 2 root root 4096 2006-11-28 23:17 /etc/rcS.d
--------------------------------------------------------------------------------
rlmanager: sys-v-init hack
~~~~~~~~~~~~~~~~~~~~~~~~~~
rlmanager: one script to be executed, sources
simply configuration
Gentoo Linux
~~~~~~~~~~~~
no content yet
BSD init
~~~~~~~~
Analyzed on FreeBSD.
/etc/rc
/etc/rc.d/ system services
/usr/local/etc/rc.d/ user installed services
Scripts may use values from /etc/rc.conf to check whether
they should be started or not.
runit
~~~~~
Three stages: Script1, dependencies, Script3
Initng
~~~~~~
no content yet
minit
~~~~~
Focus on being small. Needs libowfat.
Log support through pipe.
Problems overview
-----------------
Scripts
~~~~~~~
Scripts are slow. Startup time. Execution time. Unecessary overhead.
Sequential startup
~~~~~~~~~~~~~~~~~~
In the 'old' init systems there is no support of parallel execution.
So if one services hangs at startup, the whole system has to wait
(for instance a dhcp request). In the worst case this may result in
a non-booting system. Systems with dependencies can start the gettys
for login before, in parallel to or even after blocking services are
started. The fact, that they *are* started means you can interact with
the system, you can end looping or blocking services.
So sequential startup may not only be the reason for a pretty
slow system, but also for an unmaintainable one.
Unecessary services
~~~~~~~~~~~~~~~~~~~
Many implementations try to do whatever-is-possible on boot, trying to
cover every possible and impossible situation.
Portmap without any program that needs it.
Starting all installed services per default (Debian) vs. let the
user choose what to start (Gentoo).
Nicht entfernte Altlasten ab und zu.
No dependencies
~~~~~~~~~~~~~~~
Systems without dependencies may result in a chaos state after booting up.
Only soft dependencies
~~~~~~~~~~~~~~~~~~~~~~~
The dependencies are used, but because of soft dependencies, services
may be started although their dependency is missing.
Parallel init system debugging
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
In contrast to the old init systems, parallel executing init systems
are much harder to debug. If the starting init systems does not have
proper logging / reporting to the user (like cinit-0.2 has), the user
may never find out, what happened and why.
Parallel / dependency checking init system with variable boot order
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The boot order may be different each time you boot and is dependent
on the success or failure of services. This way you can never tell
exactly how and in which order they are executed. On multiprocessor
machines, where processes may be truely executed in parallel, there
is no order anymore, but there are n orders on n processors.
To solve this issue, init systems which support parallel execution
should ship with some testing tools, that simulate a startup.
cinit will ship with such a tool in cinit-0.4.
References
----------
no content yet