Skip to Content.
Sympa Menu

perfsonar-dev - Re: [pS-dev] Re: perfSONAR error logging

Subject: perfsonar development work

List archive

Re: [pS-dev] Re: perfSONAR error logging


Chronological Thread 
  • From: "Luchesar V. ILIEV" <>
  • To: Domenico Vicinanza <>
  • Cc: Nicolas Simar <>, Jochen Reinwand <>, , ,
  • Subject: Re: [pS-dev] Re: perfSONAR error logging
  • Date: Fri, 06 Jun 2008 18:55:37 +0300
  • Disposition-notification-to: "Luchesar V. ILIEV" <>
  • Openpgp: id=9A1FEEFF; url=https://cert.acad.bg/pgp-keys/
  • Organization: BG.ACAD/IPP-BAS

Hi Domenico,

Domenico Vicinanza wrote:
the format "perfSONAR-<SERVICE_ID>" as first word of the log entry is an important requirement for us because that is the variable syslog-ng associates to the responsible of the log. So it could not be a date or simply the location.

I'm afraid I can't understand that. Could you please elaborate about that variable and it's role for syslog-ng. (I'm not questioning your choice in any way, I'd just like to better understand what you're doing.)

The other fields can be organized as you were suggesting. Our perfSONAR logentries should start with

perfSONAR-AS ...
perfSONAR-LS ...
...

I guess today is not my day... What do AS and LS mean?

Just as a remark, we preferred for the time being to not use the log4j but adding the prefix "perfSONAR-<SERVICE_NAME>" in the syslog-ng configuration, for example, perfSONAR AS logfile:

source aslog{
pipe("/dev/aslog" log_prefix("perfSONAR-AS: ") );
};

In this way if perfSONAR services need to be updated or reinstalled we do not have to set the format preferences again.

So, either today is indeed not my day, or you I'm guessing correctly that you capture "non-syslog" log files and pipe them to syslog. Right?

I hope I'm not really asking (too) stupid questions. :)

Thanks,
Luchesar


Many thanks anyway to you and Roman.

Cheers,
Domenico


Luchesar V. ILIEV wrote:
OK, here's a modified proposal:

% LOCATION % SERVICE_ID % SEVERITY % bla-bla-bla

e.g.:

% DANTE % SSH_TELNET-MP % DEBUG % bla-bla-bla

or in full:

May 29 17:29:58 srv3 % DANTE % SSH_TELNET-MP % DEBUG % ExistDbFactory: Creating XML RPC Storage Manager

From my (humble) experience it should be quite easy parsable with perl or awk for instance (using " % ", that is % with the spaces around, as delimiter.) Still, it looks in my eyes easy to digest, and placing location before service actually seems more logical in hierarchical sense.

Comments?

Cheers,
Luchesar

P.S. Is "perfSONAR-" before SERVICE_ID required?

P.P.S. Slightly better looking, but requiring two-pass parsing could be:

%% LOCATION :: SERVICE_ID :: SEVERITY %% bla-bla-bla

e.g.:

%% DANTE :: SSH_TELNET-MP :: DEBUG %% bla-bla-bla

or in full:

May 29 17:29:58 srv3 %% DANTE :: SSH_TELNET-MP :: DEBUG %% ExistDbFactory: Creating XML RPC Storage Manager



Luchesar V. ILIEV wrote:
Domenico Vicinanza wrote:
Hi Nicolas,

exactly, in particular, if it is possible:

[perfSONAR-SERVICE_ID]
[@LOCATION]
SEVERITY% DESCRIPTION

It is important for us to have the serviceID in the first word of the log: it is the way syslog-ng is able to detect the program which caused the log and allows the proper classification in our database. For example:

perfSONAR-RRDMA srv3.ams.nl.geant2.net WARNING <description>

It is better, to ease the parsing process, not to have dots (.) as separator between location and severity, since the host names contain dots too.

That's very good point, indeed. Concerning location, however, do you really prefer it to be hostname? E.g. for BREN/ISTF, the domain name will be... something.acad.bg. Not very intuitive, I'm afraid.

Anyway, if we'd like to do fast parsing, we could use something like this:

%pS-RRD_MA%srv3.ams.nl.geant2.net%WARNING% bla-bla-bla

I, personally, wouldn't be happy using spaces as delimiters, as they are basically ubiquitous, and so make a very difficult to handle delimiter.

The percent sign was inspired by, as you might've guessed, Cisco. It's quite rare, but even then perhaps we should impose a restriction on using it in the messages/descriptions themselves. Yet, as a beginning, I think it should be OK even without.

Sure, the percent sign makes the messages less legible, but it's a balance we must find: legibility vs ease of parsing (just like with the dots, BTW.)

Cheers,
Luchesar





Attachment: signature.asc
Description: OpenPGP digital signature




Archive powered by MHonArc 2.6.16.

Top of Page