Re: [gentoo-user] Re: Debian just voted in systemd for default init system in jessie

["Yuri K. Shatroff" <yks-uno@yandex.ru>]

20.02.2014 19:24, Alan McKinnon пишет:
> On 20/02/2014 13:53, Yuri K. Shatroff wrote:
>> I don't need such 'solutions' to non-existent problems. But if there
>> were a *real* necessity to pretty-print a log's tail in service status,
>> I think it would have been a matter of a proper setup (i.e. the service
>> using syslog, hence a defined log format) and not a heck more complicated.
>>
>>> Definetly not a 5-minutes job.
>>
>> 5 minutes is even too much to type sort of
>> tail -${LINES} ${SERVICE}.log
>> if you know where to look up LINES and SERVICE.
>
>
> You've never actually tried this, right?

You probably misunderstood. I don't *intend* to try this myself with 
existing tools, I'm speaking of the init scripts modification. I say 
that this modification of e.g. OpenRC, if required, would be done quite 
easily with some assumptions.

> Your idea instantly fails as the rc-service author has no idea of what
> you defined ${SERVICE} to be and no way to determine what it is now.

Yes, the rc-service author does not have any idea because he is not 
requested to.
${SERVICE} obviously comes from `rc-service status ${SERVICE}` .
The result (e.g. tail -n {$LINES} ${SERVICE}.log) is achieved by:
1. putting LINES= in /etc/conf.d/${SERVICE}
2. setting up ${SERVICE}.log with syslog. (or putting LOGFILE=... and 
doing `tail -n ${LINES} ${LOGFILE}, or even LAST_LOG_CMD=`mysql -qe 
'SELECT ... FROM log.log ORDER BY date DESC LIMIT ${LINES}'`, or *whatever*)
3. adding this `tail -n ...` or whatever call to the init script .
4. voila.

If you feel I'm again entirely wrong please point out why.

> How are you going to deal with the situation with a big busy daemon that
> immediately starts serving requests when started (i.e. with very little
> delay)?

Either you or I seem to have misunderstood again.
The problem in question IMO was to add the output of last N log entries 
to `*service status` analogous to systemctl status.
When you do tail -n $FILE, don't you *always* get the last N lines of 
the file at the moment of issuing the cmd, regardless whether the file 
is being added a million lines per second. I don't think that journalctl 
can essentially work differently.

> By the time grep, sed, awk and friends have gotten around to making
> their way through a log file of varying size, the entries that apply to
> restart can easy be many hundreds of log lines prior.

Why do you refer to restart?

Canek wrote:

 >> systemctl status apache2.service

 >> (see [2]) will print the status of the Apache web server, and also
 >> the last lines from the logs. You can control how many lines

I don't notice anything about restart here. Just print out the last N lines.

If the question were about [re]start logs, and if in general you are 
getting millions of entries written to the logs, you could use DBMS (not 
necessarily relational). Maybe this *does* require some mess to setup 
(we did it back in times of SunOS), but it could be resolved with 
OpenRC/any SysV/BSD init (at the init-scripts level) if really necessary.
Am I wrong?

> I have done this, and it does not work. I got a result and it's
> relaible, but you don't want to know what it took. It's also highly
> customized and useless to anything other than my highly customized setup.

Well, if you have to set up one system from scratch then probably it's 
easier to use one generalized tool. But if you have an already 
long-working setup which suits your needs, I believe it's relatively 
easy to deploy it on other systems.
I don't like truisms but there is no generic setup suitable for 
everything. Neither is systemd-journald.

-- 
Regards,
Yuri K. Shatroff