On 09/15/2017 07:26 AM, Kevin A. McGrail wrote:
On 9/15/2017 8:22 AM, Merijn van den Kroonenberg wrote:
So one of the problems with solving this is because getting help requires
a complicated enrollment.

Maybe there are chunks of work which can be offloaded to the community
without requiring full sysadmin enrollment?

I can imagine lots of scripts are involved into running the rule
generation. Maybe some need reviewing or something? (just throwning thing
in the air here)

I could be forest for the trees but I haven't seen the ability for anyone to make progress with minimal access.  Dave, tell me if you agree, please?


Regards,

KAM

I hit a wall with my troubleshooting a couple of months ago. Normally I am able to solve very detailed issues like this without much problem.  Here are the issues:

1. These are very old scripts written in disjointed pieces over a long period of time by different people with different styles. There is very little consistency and no proper modularity by function.

2. It takes a long time, 40+ minutes, to do a single run making troubleshooting very slow and difficult.  The scripts are mostly using very verbose bash "set -x" output for logging which is hard to dig through the massive output and correlate with problems. They need to have proper debug logging after being rewritten in modular parts.

3. I have narrowed down the problem to the general area of a perl Makefile which builds a custom garescorer.c file which does some statistical analysis to determine the best score for rules in the 72_scores.cf.  These 72_scores.cf are excluded from 50_scores.cf (static scores) and are currently incomplete making these rules default to 1.0.  Most of theses missing rules should be much higher than 1.0 causing SA to allow spam through on most installations that don't have an optimized MTA in front of SA.

https://wiki.apache.org/spamassassin/InfraNotes2017#mkupdates

~/svn/trunk/build/mkupdates/mkupdate-with-scores

    masses -> perl Makefile.PL && make (complete build of SA and test)
        - perl hit-frequencies
        - garescorer - compiles and runs it, requires build/pga                          <--- THE PROBLEM IS NEAR HERE

I think the problem is somewhere behind line 127 and 128 which does many things/steps:

http://svn.apache.org/viewvc/spamassassin/trunk/build/mkupdates/

4. Most of the ruleqa/masscheck scripts are sh or bash.  I am not a perl person.  I am good at bash, Python, and a few others but not perl.  I can hack my way though reading and updating existing perl but that's about it.


I have been thinking about this the past few weeks and maybe there is a "band-aid" option where we could merge the current 72_scores.cf rules that are missing with the shorter version that is being generated today to keep the 72_scores.cf complete and get the rule updates happening again.

Incomplete version from last night:

http://svn.apache.org/viewvc/spamassassin/trunk/rulesrc/scores/72_scores.cf?revision=1808406&view=markup&sortby=date

Last good version:

http://svn.apache.org/viewvc/spamassassin/trunk/rulesrc/scores/72_scores.cf?revision=1786976&view=markup&sortby=date

Dave

--
David Jones

Reply via email to