Comments on: Joins

By: Joins – HJ « Oracle Scratchpad

Joins – HJ « Oracle Scratchpad — Mon, 06 Sep 2010 13:17:42 +0000

[...] Comments Joins – MJ « Oracle Scratchpad on JoinsScalability Conflict « Oracle Scratchpad on HeisenbergFragmentation 1 « Oracle [...]

By: Joins – MJ « Oracle Scratchpad

Joins – MJ « Oracle Scratchpad — Mon, 06 Sep 2010 13:12:32 +0000

[...] Joins – MJ Filed under: Execution plans,Performance — Jonathan Lewis @ 5:50 pm UTC Aug 15,2010 The final join mechanism in my “all joins are nested loop joins” argument is the Merge Join – a join mechanism that depends on both its row sources being pre-sorted on the join columns. ((For a quick reference list of URLs to all three articles in turn, see: Joins.)) [...]

By: Jonathan Lewis

Jonathan Lewis — Thu, 12 Aug 2010 07:48:09 +0000

Stefan, Thanks for your suggestions. I was just a little careful to use the term "join mechanism" in my original statement - the cluster join isn't an extra join mechanism: in fact, as you suggest in your final sentence, the commonest mechanism used in a cluster join is the nested loop. (Merge joins - cartesian, inevitably - can also appear). Example of plan: [sourcecode gutter="false"] ------------------------------------------------------------------------------- | Id | Operation | Name | Rows | Bytes | Cost | ------------------------------------------------------------------------------- | 0 | SELECT STATEMENT | | 10 | 300 | 3 | | 1 | NESTED LOOPS | | 10 | 300 | 3 | | 2 | TABLE ACCESS BY INDEX ROWID| INDEXED_TABLE1 | 1 | 15 | 2 | |* 3 | INDEX RANGE SCAN | IT_I1 | 1 | | 1 | |* 4 | TABLE ACCESS CLUSTER | INDEXED_TABLE2 | 10 | 150 | 1 | ------------------------------------------------------------------------------- Predicate Information (identified by operation id): --------------------------------------------------- 3 - access("T1"."SMALL_VC"='0000001000') 4 - filter("T2"."ID"="T1"."ID") [/sourcecode] The two tables are in the same index cluster the ID column is the cluster key and the join is on the cluster key. Note how we have a special access method (table access cluster) but, as you assumed, a standard join mechanism (nested loop).

By: Stefan Schwetschke

Stefan Schwetschke — Wed, 11 Aug 2010 09:19:18 +0000

Actually there are at least four kinds of joins used by Oracle: Merge, Hash, Nested Loops and Cluster Joins.

The latter are used when you join two tables that are stored in the same data blocks (clustering). When you join them in a way that each of the rows being joined share the same data block, Oracle can easily join them together without much additional efford.

I haven’t tested this, but I guess in the execution plan the cluster join will show up as a nested loop join with a cluster access path on all but one of the tables accessed.

By: David Aldridge

David Aldridge — Tue, 10 Aug 2010 11:52:28 +0000

Opinion 1 is closest to my own, and therefore the most likely to be correct ;) “Troubleshooting a problem” = “Troubleshooting”, or “Problem-shooting” if you will. It’s the sort of vague thing you say you’re doing when you want someone to just go away reassured that you’re doing “something” about a problem. It could mean so many things that in the end it means almost nothing.

Anyway, I’m putting “troubleshooting” on my List Of Forbiddden Words and Phrases.

By: Bix

Bix — Fri, 06 Aug 2010 14:40:52 +0000

I am trouble-shooting a problem
I was trouble-shooting a problem

i just ‘think’ the above both are wrong sentences…

I am on trouble-shooting a problem
I am good at trouble shooting

By: Bix

Bix — Fri, 06 Aug 2010 14:32:02 +0000

By definition – ‘JOIN’ means – to ‘look up’ a value (from set A ) in Set B.

the basic task is to ‘look up’ which is a task basically in all 3 types of JOINS – they way Oracle ‘organise’ and ‘arrange’ the tables before starting look-up is different – that is all about ‘start up’ cost

By: marcov

marcov — Fri, 06 Aug 2010 09:58:58 +0000

NLJ: The driver outer loop is executed at most once, while the inner loop may or not be executed at all: it can return the first rows while computing the entire result set.

MJ: each step is executed once, but before returning the first rows it must read all the inputs, because it needs to sort them

HJ: each step is executed once. Before returning the first rows only the first step must read its input completely.

By: joel garry

joel garry — Tue, 03 Aug 2010 20:44:43 +0000

Well if you are a scientist, you have to put it in the most passive third person:

The trouble-shooting of the problem was completed by the consultant.

(Slightly-less-jokey answer: it’s slang for various forms of domain-specific forensic engineering, so somewhere in the land between null and not-yet-properly-defined)

By: Flado

Flado — Tue, 03 Aug 2010 11:40:26 +0000

Any operation on a potentially infinite set must employ a loop (or recursion, but let’s not go there). A join is an operation on two potentially infinite sets where each element of one of the sets must be searched for in the other set, so you need to process a potentially infinite set for every member of another potentially infinite set. Hence, the corresponding loops must be nested. Q.E.D.
Disclaimer: I might be talking complete rubbish (wow, it’s been 20 years already since that course).

Flado