RE: Exadata Tuning Question+
- From: "Mark W. Farnham" <mwf@xxxxxxxx>
- To: <iggy_fernandez@xxxxxxxxxxx>, "'Jonathan Lewis'" <jonathan@xxxxxxxxxxxxxxxxxx>, "'ORACLE-L'" <oracle-l@xxxxxxxxxxxxx>
- Date: Tue, 11 Nov 2014 10:02:05 -0500
long thread. I may have missed a few things.
Regarding just the bit:
LEFT OUTER JOIN service_relationship sr ON sr.sid = slv.sid AND
sr.relationship_level IN ('1', '2', '3', '4', '5')
There might not be an index or partitioning on sr.relationship_level. Even
so, I wonder whether
left outer join
(select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = '1' union all
select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = '2' union all
select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = '3' union all
select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = '4' union all
select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = '5'
) sr on sr.sid = slv.sid
might be effective. I hope I didn't miss any columns needed from
service_relationship or inadvertently munge the syntax.
Presumably minimizing the returned columns and rows prior to the outer join
is useful. IS '1' thru '5' all the rows of service_relationship?
I'm not sure whether this will be effective in the overall umbrella of
producing the result set.
In the alternative (especially if '1' through '5' does project all the rows
of service_relationship, does the column minimization of:
left outer join
(select relationship_level, sid, id, related_sid from service_relationship
where relationship_level = in ('1', '2', '3', '4', '5')
) sr on sr.sid = slv.sid
help?
mwf
From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of Iggy Fernandez
Sent: Tuesday, November 11, 2014 8:22 AM
To: Jonathan Lewis; ORACLE-L
Subject: RE: Exadata Tuning Question+
Jonathan,
In plan 4110902776 posted by Abdul, the hash join at line 3 is predicted to
return 6539 thousand rows(the cardinality of SERVICE_LOOKUP) probably
because SERVICE_LOCATION_ID is the primary key of SERVICE_LOCATION and
because there probably is a foreign key relationship between SERVICE_LOOKUP
and SERVICE_LOCATION. Another guess is that there is a parent-child
relationship between SERVICE_LOOKUP (parent) and SERVICE_RELATIONSHIP
(child). Also, SERVICE_LOOKUP is not being filtered. I was therefore
expecting 21 million rows (the cardinality of SERVICE_RELATIONSHIP) from the
outer hash join at line 2.
In any case, since SERVICE_LOOKUP is not being filtered, your suggestion to
"aggregate then join" is probably the best bet for Abdul. Hopefully, Abdul
has enough sort space.
Iggy
_____
From: jonathan@xxxxxxxxxxxxxxxxxx
To: oracle-l@xxxxxxxxxxxxx
Subject: RE: Exadata Tuning Question+
Date: Tue, 11 Nov 2014 12:40:58 +0000
Iggy,
In the earlier posts we saw an estimated 25 rows from service_relationship
for each level for each sid, and at one point you corrected a multiplier of
25 to 125. The initial explosion is actually short of that estimate: in
simplistic terms we might have expected the join to take us from 7M to 875M
rows, which is why I posted the comment about considering the trade off
between "join then aggregate" and "aggregate then join". The fact that the
206M rows from the hash join are predicted to return 206M rows after
aggregation is just an unfortunate feature of the way that Oracle estimates
aggregations.
Regards
Jonathan Lewis
http://jonathanlewis.wordpress.com
@jloracle
_____
From: oracle-l-bounce@xxxxxxxxxxxxx [oracle-l-bounce@xxxxxxxxxxxxx] on
behalf of Iggy Fernandez [iggy_fernandez@xxxxxxxxxxx]
Sent: 11 November 2014 12:28
To: Ebadi, Abdul; ORACLE-L
Subject: RE: Exadata Tuning Question+
Reposting since my reply was not posted
Abdul,
I cannot understand why the number of rows exploded from 21 million to 206
million in line 2. Surely, the number of rows cannot be more 21M because SID
is (or appears to be) the primary key of SERVICE_LOOKUP and has a foreign
key relationship to with SERVICE_RELATIONSHIP. Perhaps, primary key and
foreign key constraints are not being enforced. I need to review the join
conditions in the predicate section of the query plan. Because of the rows
explosion, the required sort area size is 10G. You can try increasing the
sort_area_size to 10G. A better strategy might be to perform the group by
before the join using the WITH CLAUSE or the PLACE_GROUP_BY hint.
Could you post the following two EXPLAIN PLAN (including the predicate
sections)
EXPLAIN PLAN FOR
SELECT
slv.sid,
slv.service_location_id,
RTRIM(XMLSERIALIZE(CONTENT EXTRACT( XMLAGG(XMLELEMENT("e",
DECODE(sr.relationship_level, '1', sr.RELATED_SID || ',', NULL)) ORDER BY
sr.RELATED_SID), '//text()' )) , ',' ) AS relatedserviceinstanceidlevel1,
RTRIM(XMLSERIALIZE(CONTENT EXTRACT( XMLAGG(XMLELEMENT("e",
DECODE(sr.relationship_level, '2', sr.RELATED_SID || ',', NULL)) ORDER BY
sr.RELATED_SID), '//text()' )) , ',' ) AS relatedserviceinstanceidlevel1,
RTRIM(XMLSERIALIZE(CONTENT EXTRACT( XMLAGG(XMLELEMENT("e",
DECODE(sr.relationship_level, '3', sr.RELATED_SID || ',', NULL)) ORDER BY
sr.RELATED_SID), '//text()' )) , ',' ) AS relatedserviceinstanceidlevel1,
RTRIM(XMLSERIALIZE(CONTENT EXTRACT( XMLAGG(XMLELEMENT("e",
DECODE(sr.relationship_level, '4', sr.RELATED_SID || ',', NULL)) ORDER BY
sr.RELATED_SID), '//text()' )) , ',' ) AS relatedserviceinstanceidlevel1,
RTRIM(XMLSERIALIZE(CONTENT EXTRACT( XMLAGG(XMLELEMENT("e",
DECODE(sr.relationship_level, '5', sr.RELATED_SID || ',', NULL)) ORDER BY
sr.RELATED_SID), '//text()' )) , ',' ) AS relatedserviceinstanceidlevel1
FROM
service_lookup slv
LEFT OUTER JOIN service_location sl ON sl.service_location_id =
slv.service_location_id
LEFT OUTER JOIN service_relationship sr ON sr.sid = slv.sid AND
sr.relationship_level IN ('1', '2', '3', '4', '5')
GROUP BY slv.id, slv.service_location_id;
EXPLAIN PLAN FOR
WITH sr AS (
SELECT
sr.id,
LISTAGG(DECODE(sr.relationship_level, '1', sr.related_sid, NULL) ',')
WITHIN GROUP (ORDER BY sr.related_sid) AS relatedserviceinstanceidlevel1,
LISTAGG(DECODE(sr.relationship_level, '2', sr.related_sid, NULL) ',')
WITHIN GROUP (ORDER BY sr.related_sid) AS relatedserviceinstanceidlevel2,
LISTAGG(DECODE(sr.relationship_level, '3', sr.related_sid, NULL) ',')
WITHIN GROUP (ORDER BY sr.related_sid) AS relatedserviceinstanceidlevel3,
LISTAGG(DECODE(sr.relationship_level, '4', sr.related_sid, NULL) ',')
WITHIN GROUP (ORDER BY sr.related_sid) AS relatedserviceinstanceidlevel4,
LISTAGG(DECODE(sr.relationship_level, '5', sr.related_sid, NULL) ',')
WITHIN GROUP (ORDER BY sr.related_sid) AS relatedserviceinstanceidlevel5
WHERE sr.relationship_level IN ('1', '2', '3', '4', '5')
GROUP BY sr.id
)
SELECT
slv.sid,
slv.service_location_id,
sr.relatedserviceinstanceidlevel1,
sr.relatedserviceinstanceidlevel2,
sr.relatedserviceinstanceidlevel3,
sr.relatedserviceinstanceidlevel4,
sr.relatedserviceinstanceidlevel5
FROM
service_lookup slv
LEFT OUTER JOIN service_location sl ON sl.service_location_id =
slv.service_location_id
LEFT OUTER JOIN sr ON sr.sid = slv.sid;
From: Abdul.Ebadi@xxxxxxxxxx
Thanks for the reply Iggy.
Your second suggestion below runs faster, but according to our developer we
"cannot use LISTAGG as the string is larger than 4000 chars for some of the
values, so will have to use XMLAGG"
The first suggestion is OK and missing a group by, but when we added group
by it didn't finish even after long time.
Here is the plan for your second option (even switching to SORT MERGE join
resulted in same thing - not finished):
Plan hash value: 4110902776
----------------------------------------------------------------------------
---------------------------------
| Id | Operation | Name | Rows | Bytes
|TempSpc| Cost (%CPU)| Time |
----------------------------------------------------------------------------
---------------------------------
| 0 | SELECT STATEMENT | | 206M|
9067M| | 151K (1)| 00:00:06 |
| 1 | SORT GROUP BY | | 206M|
9067M| 10G| 151K (1)| 00:00:06 |
|* 2 | HASH JOIN OUTER | | 206M|
9067M| | 2311 (7)| 00:00:01 |
|* 3 | HASH JOIN | | 6539K|
149M| | 1864 (4)| 00:00:01 |
| 4 | TABLE ACCESS STORAGE FULL| SERVICE_LOCATION | 3175K|
18M| | 357 (1)| 00:00:01 |
|* 5 | TABLE ACCESS STORAGE FULL| SERVICE_LOOKUP | 6539K|
112M| | 1503 (4)| 00:00:01 |
PLAN_TABLE_OUTPUT
----------------------------------------------------------------------------
----------------------------------------------------------------------------
------------------------------------------------
|* 6 | TABLE ACCESS STORAGE FULL | SERVICE_RELATIONSHIP | 21M|
454M| | 376 (8)| 00:00:01 |
----------------------------------------------------------------------------
---------------------------------
Thanks,
Abdul
Other related posts:
