Re: single instance of multilog for multiple processes

On Mon, 21 Sep 2009 22:34:04 +0930
Daryl Tester <[email protected]> wrote:

> > "man fifo" is not clear enough about this matter. The question is:
> > where is the limit? Is it a block of 4096 bytes?
> 
> There are two limits as such - one is PIPE_BUF, which under Linux
> (well, the one I'm running) is defined in <linux/limits.h> as 4096
> characters.  The other limit is more of an issue than limit - the
> write() to pipe has to be atomic.  In my aforementioned issue I had
> some concurrent shell scripts being supervised.  Despite preformatting
> the print strings to echo as one entire quoted string, strace showed
> bash was splitting up the writes and thus allowing other log messages
> to get interleaved.  I solved the problem by not using bash :-/.

Ok, so a few "run" scripts containing:

exec 1>/path/to/fifo
exec 2>/path/to/fifo

should work well, depending on the "atomicness" of the writing
processes.

> > I just write lines no longer than let's say 80 to 100 charachters
> > (including the timestamp) so I presume there will be no problem.
> 
> So long as you write it in one operation then it shouldn't be
> an issue  e.g. if under C -
> 
> /* Note cunning ignorance of return value */
> write(1, "hi there\n", strlen("hi there\n"));
> 
> If you split the string into several writes, that's when the
> opportunity for mischief arises.

ack

> > And how would you merge two outputs from two separate multilog
> > processes into file. Using a postprocessor?
> 
> If you have timestamps, then man (1) sort is your friend.

I know. But "tail -F current" doesn't work. I use these logs not for
"watching what has happened" but for "follow what happens". And as
there are 4 processes involved I'd like to keep them together in 1 log.
You can ask "why not write to syslog?". Well, in that case I would never
have subscribed to this list ;-)

> > I could also use "setlock" to lock the fifo for writing.
> 
> That would work, but then it wouldn't be concurrent ...

But finally it resolves the problem ;-) But OTOH, not in case of an
"exec 1>fifo" construction :(

> > And 
> 
> And?  Therefore?  Nevertheless?  But?  :-)

oops... :-)

-- 
___________________________________________________________________
It is better to remain silent and be thought a fool, than to speak
aloud and remove all doubt.

+------------------------------------------------------------------+
| Richard Lucassen, Utrecht                                        |
| Public key and email address:                                    |
| http://www.lucassen.org/mail-pubkey.html                         |
+------------------------------------------------------------------+