Re: Correlating Asterisk CDRs

On 12/07/2011 16:14, Raj Mathur (राज माथुर) wrote:
> On Wednesday 07 Dec 2011, Julien Cigar wrote:
>> Try to raise work_mem
>
> Odd, I tried adding work_mem=50MB / 256MB / 1024MB into postgres.conf
> and the times actually went up to over 12 seconds.  Leaving it commented
> results in the 4-second time originally posted.

sorry I replied too fast.. There is no external disk merge so increasing
work_mem is useless. Problem is the "merge join". How big is the table?
Are the statistics up to date?

>
> Regards,
>
> -- Raj
>
>> On 12/07/2011 15:34, Raj Mathur (राज माथुर) wrote:
>>> I'm trying to correlate Call Data Records (CDRs) from two Asterisk
>>> servers, one of which uses the other for telephony.  The data is in
>>> the tables cdr and cdr2.  With some indexes, the query and explain
>>> result are:
>>>
>>> explain analyse select cdr.calldate, cdr2.calldate,
>>> (cdr2.calldate-cdr.calldate) as rtdur, cdr.clid, cdr.dst, cdr2.src,
>>> cdr2.dst, cdr2.dstchannel, cdr2.lastapp, cdr2.duration,
>>> cdr2.disposition from cdr, cdr2 where cdr2.calldate>= cdr.calldate
>>> and cdr.clid=cdr2.clid and cdr.dst=substring(cdr2.dst from 4) order
>>> by cdr.calldate, cdr2.calldate, cdr.clid limit 100;
>>>
>>>                                                                QUERY
> PLAN
>>>
>>> -------------------------------------------------------------------
>>> -------------------------------------------------------------------
>>> -------
>>>
>>>    Limit  (cost=46782.15..46782.40 rows=100 width=109) (actual
>>>    time=4077.866..4078.054
>>>
>>> rows=100 loops=1)
>>>
>>>      ->    Sort  (cost=46782.15..46785.33 rows=1272 width=109)
>>>      (actual time=4077.863..4077.926
>>>
>>> rows=100 loops=1)
>>>
>>>            Sort Key: cdr.calldate, cdr2.calldate, cdr.clid
>>>            Sort Method:  top-N heapsort  Memory: 42kB
>>>            ->    Merge Join  (cost=2.95..46733.54 rows=1272
>>>            width=109) (actual
>>>
>>> time=0.070..3799.546 rows=168307 loops=1)
>>>
>>>                  Merge Cond: (((cdr.clid)::text = (cdr2.clid)::text)
>>>                  AND ((cdr.dst)::text =
>>>
>>> "substring"((cdr2.dst)::text, 4)))
>>>
>>>                  Join Filter: (cdr2.calldate>= cdr.calldate)
>>>                  ->    Index Scan using ick1 on cdr
>>>                  (cost=0.00..34667.86 rows=208798
>>>
>>> width=43) (actual time=0.022..434.246 rows=208798 loops=1)
>>>
>>>                  ->    Index Scan using i2k1 on cdr2
>>>                  (cost=0.00..9960.89 rows=65449 width=88)
>>>
>>> (actual time=0.011..391.599 rows=240981 loops=1)
>>>
>>>    Total runtime: 4078.184 ms
>>>
>>> (10 rows)
>>>
>>> Is there any way to make this query faster?  I already have an
>>> index i2k1 on substring(cdr2.dst from 4), which is being used.
>>>
>>> Application
>>> -----------
>>>
>>> I'm looking for all caller records in cdr2 that have the same
>>> callerid (clid) and destination (dst) and were started on cdr2
>>> after they were started on cdr.  cdr2.dst is the same as cdr.dst
>>> but with a 3-character prefix.
>


--
No trees were killed in the creation of this message.
However, many electrons were terribly inconvenienced.