Re: [spamdyke-users] Question about Greylisting and deleting Zero-Length-Entries

2011-11-02 Thread trog
Quoting Eric Shubert e...@shubes.net:

 I've been wondering though about perhaps using tmpfs for the graylist
 tree. That might be a potential solution as well for hosts that process
 huge amounts of email. Of course the whole tree would be lost on
 rebooting, but if that was a problem it could be copied off periodically
 and restored. If I get some time one day, I may do some test comparisons.


The thought of using up RAM for the graylist data doesn't fit well  
with me. I'd much rather have the RAM used as file cache, for both the  
mail itself, and for things like AV signatures.

-trog

___
spamdyke-users mailing list
spamdyke-users@spamdyke.org
http://www.spamdyke.org/mailman/listinfo/spamdyke-users


Re: [spamdyke-users] Question about Greylisting and deleting Zero-Length-Entries

2011-11-02 Thread Eric Shubert
On 11/02/2011 03:11 AM, t...@uncon.org wrote:
 Quoting Eric Shuberte...@shubes.net:

 I've been wondering though about perhaps using tmpfs for the graylist
 tree. That might be a potential solution as well for hosts that process
 huge amounts of email. Of course the whole tree would be lost on
 rebooting, but if that was a problem it could be copied off periodically
 and restored. If I get some time one day, I may do some test comparisons.


 The thought of using up RAM for the graylist data doesn't fit well
 with me. I'd much rather have the RAM used as file cache, for both the
 mail itself, and for things like AV signatures.

 -trog

Me too, but it depends on the amount. We're only talking inodes really. 
Might not take up all that much space. You're running a huge amount of 
messages though, so it might be a significant amount. Just a thought.

-- 
-Eric 'shubes'

___
spamdyke-users mailing list
spamdyke-users@spamdyke.org
http://www.spamdyke.org/mailman/listinfo/spamdyke-users


Re: [spamdyke-users] Question about Greylisting and deleting Zero-Length-Entries

2011-11-01 Thread trog
Quoting Peter Palmreuther li...@zentrumderarbeit.org:


 I do exactly this (albeit with a self written script, because when I  
 started to cleanup my graylisting directory I didn't know about  
 'qtp-prune-graylist').
 Empty files older than 24 hours, too old files (-  
 graylist-max-secs) and subsequently empty directories are removed.  
 In my case once a day, when there's lower load on the server. The  
 (for me also separate) graylist filesystem offers enough space and  
 inodes to cover it's uncleaned usage for several days, so there's no  
 significant profit to gain cleaning up more than once a day.


Just my opinion, but I think you're not getting the most out of  
graylisting if you are pruning the records so aggressively, due to the  
need for the remote server to do a re-send more than necessary, the  
annoyance to users of the delays in email and there are still broken  
servers out there.

I'd be interested to see the numbers on your message count and  
graylist filesystem usage.

Anyway, here's what I do:

o I use a loopback filesystem (XFS) to hold the graylist data. This  
has two advantages, I can easily dump the entire dataset by  
re-formatting the virtual filesystem, and XFS allocates inodes  
dynamically, so won't easily run out.

o I keep 3 weeks worth of graylist history. In order to manage that,  
I've patched the code to change the layout on disk, into per-week  
directory structures (by week number), like this:
/graylist-dir/domain/week-no/recip/domain/sender

Graylist entries are then automatically migrated from the previous to  
the current weeks data when they are used by the spamdyke process.

This means that at the end of the week, I can simply delete the  
directory containing the oldest data, without having to perform any  
kind of filesystem 'find' operation looking for data that is too old,  
which is very expensive.

Thanks,
-trog
___
spamdyke-users mailing list
spamdyke-users@spamdyke.org
http://www.spamdyke.org/mailman/listinfo/spamdyke-users


Re: [spamdyke-users] Question about Greylisting and deleting Zero-Length-Entries

2011-11-01 Thread Peter Palmreuther
On 01.11.2011 at 12:03 t...@uncon.org wrote:
 
 Quoting Peter Palmreuther li...@zentrumderarbeit.org:
 I do exactly this (albeit with a self written script, because when I  
 started to cleanup my graylisting directory I didn't know about  
 'qtp-prune-graylist').
 Empty files older than 24 hours, too old files (-  
 graylist-max-secs) and subsequently empty directories are removed.  
 In my case once a day, when there's lower load on the server. The  
 (for me also separate) graylist filesystem offers enough space and  
 inodes to cover it's uncleaned usage for several days, so there's no  
 significant profit to gain cleaning up more than once a day.
 
 Just my opinion, but I think you're not getting the most out of  
 graylisting if you are pruning the records so aggressively, due to the  
 need for the remote server to do a re-send more than necessary, the  
 annoyance to users of the delays in email and there are still broken  
 servers out there.

I'd agree, if only I could see the aggressiveness ...

I have a 'graylist-max-secs' value equivalent to one week. So somebody sending 
another mail within one week resets the timeout, doesn't it?
Anybody else ... Well, there has to be some limit. Haven't had any problem with 
one week yet, but would increase if I'd see some arising.

Deleting empty files, older than 24 hours, I don't see as an aggressive act 
too. If a mail server didn't retried within 24 hours, how long am I supposed to 
wait, for not being aggressive?

Deleting empty folders I only see as a final clean up. I don't see a point in 
keeping them. But I'm open for arguments on this ...

 I'd be interested to see the numbers on your message count and  
 graylist filesystem usage.

As it's a rather small server I don't have any significant statistical data ... 
Sorry.

 Anyway, here's what I do:
 
 o I use a loopback filesystem (XFS) to hold the graylist data. This  
 has two advantages, I can easily dump the entire dataset by  
 re-formatting the virtual filesystem, and XFS allocates inodes  
 dynamically, so won't easily run out.

So do I. I just have not done the cleanup from the very beginning and ran out 
of inodes once.
That's why I search for cleanable data every day ... Which does, as explained 
above, does not imply I'm deleting everything older than one day ;-)

 o I keep 3 weeks worth of graylist history.

I'm not that interested in graylist history yet, that's why I don't keep stuff 
that long or analyze systematically.

 In order to manage that,  
 I've patched the code to change the layout on disk, into per-week  
 directory structures (by week number), like this:
 /graylist-dir/domain/week-no/recip/domain/sender

Well ... This patch might in fact be interesting, because:

 This means that at the end of the week, I can simply delete the  
 directory containing the oldest data, without having to perform any  
 kind of filesystem 'find' operation looking for data that is too old,  
 which is very expensive.

Right. And exactly for this very reason I do the daily run, because it simply 
reduces the number of existing file system entries more steady than once a 
week and therefore makes the 'find' a little less expensive.
Having weekly folders I'd most probably switch to delete the whole (too old) 
week too.
As 'find' would not be necessary in that situation the number of file system 
accesses should, in sum, be the same, wether I delete day-by-day or just the 
whole weekly parent directory.

So ... thanks for the dialogue. New perspectives, different opinions and some 
input keeps the topic vivacious.
Interested in the patch (maybe even it's making it's way upstream?) ...
-- 
Best regards,

Peter
___
spamdyke-users mailing list
spamdyke-users@spamdyke.org
http://www.spamdyke.org/mailman/listinfo/spamdyke-users