[Dovecot] Clarifications on Pigeonhole and MySQL lookups

[Sandro Tosi]

Hi Stephan,
thanks a lot for your reply.

On 10/11/2012 10:35 PM, Stephan Bosch wrote:
> On 10/10/2012 11:23 AM, Sandro Tosi wrote:
>> Hello,
>> we're scouting if it's possible to use Pigeonhole (currently v0.3.1,
>> as this will be provided with an upcoming Debian package) with MySQL
>> dict lookups with the mail setup we're designing.
>>
>> Our (main) goals are:
>>
>> 1. store the filters on the database
> That is possible with some limitations.

Are the ones below the only limitatios (ie one script per user) or are 
there any other worth knowing?

>> 2. allow each user to enable/disable any of the filters set we provide
>> (it's a static set of some general filters, available to all the
>> users; we're currently not providing the possibility to users to write
>> their own filters)
> Will one or multiple scripts be active at the same time?

Yep, the idea is that any user could have multiple scripts active at the 
same time, and we'd like also to give them an ordering, so like managing 
a sort of priority (the lower the priority the sooner the script is 
executed, or the other way around, doesn't matter).

Ideally, we have a set of several scripts and each user can select to 
enable only some of them, and choose the order of their executions.

>> For point 1) we already see[1] that's possible, but it uses the map
>> construct that might not fit with our current database structure: we
>> have a domain table (storing the domain info) and a mailbox table
>> (storing the mailbox info, but the username is composed by the local
>> part, stored in this table, and the domain part is a FK to the domain
>> table, using an id).
>>
>> Do you think it's possible to run a join query on domain+mailbox to
>> retrieve the mailbox_id needed to query the table for the filters? Or
>> do we have to create the filter table and store the local at domain.ext
>> info there ("relaxing" the integrity relationships between tables)?
>
> My SQL is a bit rusty, but afaik this is possible with a JOIN or a
> nested query.

Ah no well, I mean, using map { } constructs :)

The example for Sieve-MySQL only shows 2 maps, but given we've never 
used them, we'd want to know if a "map cascade" would work, so 
implementing the joins in multiple steps: selecting the ids with a map 
and the subsequent would use that id to exec the join and so on.

>> How do we specify which filters are enabled for any given user? We
>> originally thought of an "Enabled" field on the filter table, but in
>> the example in the doc[1] I hadn't seen a way to do that: it seems
>> like the filter list is specified in the proxy definition - am I
>> wrong? How can we do that?
>
> The above suggests that you would like to activate multiple Sieve
> scripts at the same time. That is currently not possible with the dict
> Script location. It is on my TODO list, but I am not sure when it will
> be ready (definitely not for coming Debian stable).

I see, I think that some others would wonder the same, so you might also 
want to extend the doc to state that explicitly.

Maybe you may want to include something in your TODO list to handle the 
ordering in case of multiple scripts.

In our situation, what would you suggest? We're now thinking of keeping 
the scripts list on a separate table, and merge the "user selected ones" 
in a single script to write in the filters table. Is that what would you 
suggest? Is there a better solution?

Cheers,
-- 
Sandro Tosi
Product Engineer
Shared Hosting Products
R&D | Dada.pro
eml sandro.tosi at register.it