Re: /lib/systemd/systemd-hostnamed hang

[Zenaan Harkness <zen@freedbms.net>]

On 11/25/12, Michael Biebl <biebl@debian.org> wrote:
> On 24.11.2012 14:40, Tom H wrote:
>> On Sat, Nov 24, 2012 at 6:02 AM, Zenaan Harkness <zen@freedbms.net>
>> wrote:
>>>
>>> Any idea how to make use of systemd-hostnamed?
>>>
>>> Eg:
>>> $ sudo /lib/systemd/systemd-hostnamed
>>> Warning: nss-myhostname is not installed. Changing the local hostname
>>> might make it unresolveable. Please install nss-myhostname!
>>> # hang's at this point, apparently indefinitely...
>>
>> What are you expecting it to do?
>
> It doesn't hang. It is a system daemon which just waits sits there and
> waits for requests (via D-Bus).
>
> Nothing unexpected here aside from starting this tool directly.
>
> It would be like starting apache by hand and then wondering that it sits
> there waiting for requests via port 80.
>
> Michael

Thanks for the clarifications.

$ lighttpd --help
lighttpd: invalid option -- '-'
lighttpd/1.4.31 (ssl) (Jul 14 2012 12:10:48) - a light and fast webserver
usage:
 -f <name>  filename of the config-file
 -m <name>  module directory (default: /usr/lib/lighttpd)
 -p         print the parsed config-file in internal form, and exit
 -t         test the config-file, and exit
 -D         don't go to background (default: go to background)
 -v         show version
 -V         show compile-time features
 -h         show this help

$ /lib/systemd/systemd-hostnamed --help
This program takes no arguments.

$ man lighttpd
#...

$ man systemd-hostnamed
No manual entry for systemd-hostnamed

I guess, being a "boot time" program and possibly headed for
inclusion-in-initrd territory, the intent is to keep "core" daemons as
small as possible, and not encumbered with memory consuming --help
option?

For discoverability/learnability of such things, should I be saying
"the source code is the man page"?

Or should I simply ignore all binaries in /lib/ ?

>From /usr/share/doc/systemd/README, :
"When systemd-hostnamed is used it is strongly recommended to
install nss-myhostname to ensure that in a world of
dynamically changing hostnames the hostname stays resolveable
under all circumstances. In fact, systemd-hostnamed will warn
if nss-myhostname is not installed. Packagers are encouraged to
add a dependency on nss-myhostname to the package that
includes systemd-hostnamed."

Perhaps the lack of libnss causes the big delays when starting xterm
after changing hostname?

Ought there be an addition to man hostname, suggesting that updating
hostname with /bin/hostname, and/ or with /etc/hostname edit, should
be accompanied with an address resolvability update to /etc/hosts in
order to avoid long timeouts when starting applications (perhaps just
under X?) ??

apropos hostname
man 5 hostname
# still no suggestion that changing hostname may cause resolver delays
man 7 hostname
# this possibly gives a bit of a hint
(please note, I totally forgot about apropos until just now)

What I am getting at here is the issue of (lack of) discoverability
regarding this hostname-causing-application-startup-delays issue and
how to fix it.

debian-user is a great fallback. It ought to be possible, when someone
in the future asks a similar question, that we can point the newbie to
say

  man hostname
  # and read paragraph 3

or say:

  apropos hostname
  man 5 hostname
  # and read paragraph 3

or some such.

I attempted to discover/solve my problem/annoyance (10+ seconds xterm
startup delays), but was not able to solve the problem with the
references to the systems documentation which I tried, and I don't
seem to see any clarity on this from man 5 or man 7 either.

I am very grateful for the abundant documentation we have, and would
like to see a little more, if that is a good idea in this case.

A rough initial draft for an extra paragraph in man 1 hostname or man
5 hostname:

WARNING (or CAUTION, or under NOTES ?)
 Changing the hostname during runtime, without the new hostname being
 resolvable to an ip address can cause delays when starting applications
 which are in any way linked to the resolver(5).
 This is the case for most(all?) X applications

Suggestions appreciated, and when we have satisfactory wording, we can
forward the suggestion to hostname-devel@lists.alioth.debian.org, and/
or create a bug against hostname package? Guidance appreciated.

Best regaqrds
Zenaan