[erlang-questions] Re: implementing a notification list

[Martin DeMello]

Thanks, Steve. When searching I need to filter on the event type as
well as retrieving records within a range of timestamps, so I'll
probably have the timestamp as the pk and index on the event type.
What worried me was this post:

--------------------------
When we scaled this up, we discovered that we were doing
one or both of the following alot: 1) insert at the front or 2) delete
from the back.  Since the underlying implementation is trying to be a
balanced tree, this leads to alot of rebalancing, which slows things down
(doubly so since our ordered set was on disk). We switched to specialized
data structures at that point.
--------------------------

It seems to me that I'd be doing much the same thing, that is
inserting a series of timestamps in increasing order, which would
potentially lead to a lot of rebalancing of the tree. I don't want to
optimise prematurely, of course, but I also don't want to find out a
few months down the line that I've made the wrong decision and have to
face a potentially expensive ripping-out and reworking.

martin

On Thu, Mar 18, 2010 at 9:07 PM, Steve Davis
<steven.charles.davis@REDACTED> wrote:
> HI Martin,
>
> A couple of approaches to consider for a starting point:
>
> 1) If appropriate you could use the timestamp as the key on your
> db_event record, i.e. -record(db_event, {timestamp, data, ...})
> 2) If using the timestamp as the key isn't appropriate, you can index
> the timestamp field, http://erlang.org/doc/apps/mnesia/Mnesia_chap5.html#id2273564
>
> hth,
> /s
>
> On Mar 18, 6:34 am, Martin DeMello <martindeme...@REDACTED> wrote:
>> I'm new to Erlang, and having a bit of trouble coming up with the
>> right way to do this:
>>
>> I have a set of mnesia triggers that watch for various db events. I
>> now want to have a persistent notification list that supports the
>> following two operations efficiently:
>>
>> 1. insert a new event, with timestamp
>> 2. retrieve all events between two given timestamps
>>
>> and possibly less efficiently
>>
>> 3. compact the table (combine several updates to a single object into
>> a single row)
>>
>> This posthttp://groups.google.com/group/erlang-programming/browse_thread/threa...
>> seems to imply that simply storing the notifications as an mnesia
>> table is a bad idea.
>>
>> martin