It would be really nice to have a setting like sa_debug_sampling (or
sa_debug_ratio, sa_debug_sample, etc.) taking an integer value (eg. 100) that
would trigger debug logging for 1 email message every sa_debug_sampling
messages (eg. 1 message every 100 messages).
The sampling can be fuzzy, for a sa_debug_sampling value of 100 I would not
mind getting a debug output every 90 to 110 messages.
Another way to address that could be time-based with a setting like
sa_debug_interval taking an integer value in seconds (eg. 10) that would
trigger debug logging for 1 email message every sa_debug_interval seconds (eg.
1 message every 10 secondes).
The implementation of such feature might have an impact on performance
even when it is disabled. The $sa_debug level is used inside
SpamAssassin to decide when to callback registered logging methods.
Without monkeypatching SA, Amavis would have to run SA with enabled
debugging all the time and filter-out dbg-loglines for non-sample
messages afterwards.
With amavis 2.13 you can prototype it yourself. Place a modified version
of Amavislog.pm inside a high-priority @INC directory, e.g:
# diff -u0 /usr/share/perl5/Mail/SpamAssassin/Logger/Amavislog.pm
/etc/perl/Mail/SpamAssassin/Logger/Amavislog.pm
--- /usr/share/perl5/Mail/SpamAssassin/Logger/Amavislog.pm 2023-05-12
00:52:23.000000000 +0200
+++ /etc/perl/Mail/SpamAssassin/Logger/Amavislog.pm 2023-06-26
18:45:41.114752331 +0200
@@ -24 +23,0 @@
- if ($args{debug}) { for (keys %llmap) { $llmap{$_} = 1 if
$llmap{$_} > 1 } }
@@ -32 +31 @@
- my $ll = $self->{llmap}->{$level};
+ my $ll = $self->is_sample(4) ? 1 : $self->{llmap}{$level};
@@ -37,0 +37,8 @@
+# sampling per amavis child
+sub is_sample {
+ my ($self, $every_nth) = @_;
+ return unless $Amavis::MSGINFO;
+ my $log_id = $Amavis::MSGINFO->log_id;
+ $self->{samples}{$log_id} = 1;
+ return %{$self->{samples}} % $every_nth == 0;
+}
