On 01/16/2015 05:05 PM, Levi Pearson wrote:
> I think a lot of people have become used to the "quirks" that surround
> the issues that other system management solutions struggle with, and
> just aren't aware of the huge variety of uses that Linux is being put
> to today and the challenges that they present to system management and
> coordination. Keeping a modern Linux system configured correctly is no
> longer just an "init" task; it's a full-time job, with dynamic
> configuration changes happening all the time.  The traditional init
> system surely doesn't cut it, and I think systemd's architecture is
> far better equipped to handle things cleanly than any of its current
> competitors.

Very well put, thank you.  I agree with you.  I don't admin a lot of
systems anymore, but I certainly encountered some of the race conditions
you describe.  In fact in RH 5, there was once a bug in nss_ldap that
would cause the machine not to boot if the LDAP server was hosted on the
same machine.  Normally I'd recommend against having LDAP running on the
same machine where nss_ldap is required, say for samba, but in a small
office sometimes you have a lot of things running on one server.  I
encountered boot-stopping things enough that I typically would edit
/etc/inittab and start a getty right at the beginning of the init
process just so I could always log in and fix things if they got stuck.

That and writing init scripts.  Yuck.  So fragile, so much re-inventing
the wheel.   On the LAS show, Poettering talked about how they looked at
a lot of common daemons on Linux, and many of them didn't quite
implement daemonizing completely correctly.  So systemd reduces the
complexity there a lot, and places daemonizing code in a place where it
can be verified and audited.  And I would wager that RH's engineers have
gone over systemd code with a fine-tooth comb.

For daemons I write myself, I think I'd much rather create a little
service file and let systemd make it into a daemon.

It's exciting to see where Linux could go in the future.  Maybe shedding
a few slashdotters to *BSDs isn't a bad thing.  But I think BSD will
have to grapple with these issues eventually if they want to continue
moving forward.  Commercial Unix certainly is trying to (SMF, launchd, etc).

Michael





/*
PLUG: http://plug.org, #utah on irc.freenode.net
Unsubscribe: http://plug.org/mailman/options/plug
Don't fear the penguin.
*/

Reply via email to