the postgresql query planner_longer

8/3/2019 The PostgreSQL Query Planner_longer

1/32

The PostgreSQL

Query PlannerRobert Haas

PostgreSQL East 2010


2/32

Why Does My Query Need

a Plan? SQL is a declarative language.

In other words, a SQL query is not a program.

No control flow statements (e.g. for, while) and

no way to control order of operations.

SQL describes results, not process.


3/32

Why Didn't The Planner Do

It My Way? Maybe your way is actually slower, or

Maybe you gave the planner bad information, or Maybe the query planner really did goof.

Related question:

How do I force the planner to use my index?


4/32

Query Planning

Make queries run fast.

Minimize disk I/O.

Prefer sequential I/O to random I/O. Minimize CPU processing.

Don't use too much memory in the process.

Deliver correct results.


5/32

Query Planner Decisions

Access strategy for each table.

Sequential Scan, Index Scan, Bitmap Index Scan.

Join strategy. Join order.

Join strategy: nested loop, merge join, hash join.

Inner vs. outer. Aggregation strategy.

Plain, sorted, hashed.


6/32

Table Access Strategies

Sequential Scan (Seq Scan)

Read every row in the table.

Index Scan or Bitmap Index Scan Read only part of the table by using the index to

skip uninteresting parts.

Index scan reads index and table in alternation.

Bitmap index scan reads index first, populatingbitmap, and then reads table in sequential order.


7/32

Sequential Scan

Always works no need to create indices inadvance.

Doesn't require reading the index, which has bothI/O and CPU cost.

Best way to access very small tables.

Usually the best way to access all or nearly therows in a table.


8/32

Index Scan

Potentially huge performance gain when readingonly a small fraction of rows in a large table.

Only table access method that can return rows insorted order very useful in combination withLIMIT.

Random I/O against base table!


9/32

Bitmap Index Scan

Scans all index rows before examining base table,populating a TID bitmap.

Table I/O is sequential, with skips; results inphysical order.

Can efficiently combine data multiple indices TID bitmap can handle boolean AND and OR

operations.

Handles LIMIT poorly.


10/32

Join Planning

Fixing the join order and join strategy is the hardpart of query planning.

# of possibilities grows exponentially withnumber of tables.

When search space is small, planner does a nearlyexhaustive search.

When search space is too large, planner usesheuristics or GEQO to limit planning time andmemory usage.


11/32

Join Strategies

Nested loop.

Nested loop with inner index-scan.

Merge join. Hash join.

Each join strategy takes an outer relation and aninner relation and produces a result relation.


12/32

Nested Loop Pseudocode

for (each outer tuple)for (each inner tuple)if (join condition is met)emit result row;

Outer or inner loop could be scanning output of

some other join, or a base table. Base table scancould be using an index.

Cost is roughly proportional to product of tablesizes bad if BOTH are large.


13/32

Nested Loop Example #1

SELECT * FROM foo, bar WHERE foo.x = bar.x

Nested LoopJoin Filter: (foo.x = bar.x)-> Seq Scan on bar-> Materialize

-> Seq Scan on foo

This might be very slow!


14/32

Nested Loop Example #2


Nested Loop-> Seq Scan on foo-> Index Scan using bar_pkey on bar

Index Cond: (bar.x = foo.x)

Nested loop with inner index-scan! Much better...though probably still not the best plan.


15/32

Merge Join

Only handles equality joins something like a.x= b.x.

Put both input relations into sorted order (usingsort or index scan) and scan through the two inparallel, matching up equal values.

Normally visits each input tuple only once, but

may need to rescan portions of the inner inputif there are duplicate values in the outer input.

Take OUTER={1 2 2 3} and INNER={2 2 3 4}


16/32

Merge Join Example

SELECT * FROM foo, bar WHERE foo.x = bar.xMerge Join

Merge Cond: (foo.x = bar.x)-> Sort

Sort Key: foo.x-> Seq Scan on foo

-> Materialize-> Sort

Sort Key: bar.x-> Seq Scan on bar


17/32

Hash Join

Like merge join, only handles equality joins.

Hash each row from the inner relation to create a

hash table. Then, hash each row from the outerrelation and probe the hash table for matches.

Very fast but requires enough memory to storeinner tuples. Can get around this using multiple

batches. Not guaranteed to retain input ordering.


18/32

Hash Join Example


Hash JoinHash Cond: (foo.x = bar.x)-> Seq Scan on foo-> Hash

-> Seq Scan on bar


19/32

Join Removal

Upcoming 9.0 feature.

Consider this query:

SELECT p.id, p.name FROM projects pLEFT JOIN person pm

ON p.project_manager_id = pm.id;

If there is a unique index on person (id), then thejoin need not be performed at all.

Common scenario when using views.


20/32

Join Reordering

SELECT * FROM fooJOIN bar ON foo.x = bar.xJOIN baz ON foo.y = baz.y

SELECT * FROM fooJOIN baz ON foo.y = baz.yJOIN bar ON foo.x = bar.x

SELECT * FROM fooJOIN (bar JOIN baz ON true)

ON foo.x = bar.x AND foo.y = baz.y


21/32

EXPLAIN Estimates

Hash Join (cost=8.28..404.52 rows=9000 width=118)Hash Cond: (foo.x = bar.x)-> Hash Join (cost=3.02..275.52 rows=9000 width=12)

Hash Cond: (foo.y = baz.y)-> Seq Scan on foo (cost=0.00..145.00 rows=10000 width=8)-> Hash (cost=1.90..1.90 rows=90 width=4)

-> Seq Scan on baz (cost=0.00..1.90 rows=90 width=4)-> Hash (cost=4.00..4.00 rows=100 width=106)

-> Seq Scan on bar (cost=0.00..4.00 rows=100 width=106)


22/32

EXPLAIN ANALYZE

Hash Join (cost=8.28..404.52 rows=9000 width=118)(actual time=0.743..51.582 rows=9000 loops=1)Hash Cond: (foo.x = bar.x)-> Hash Join (cost=3.02..275.52 rows=9000 width=12)

(actual time=0.368..30.964 rows=9000 loops=1)Hash Cond: (foo.y = baz.y)-> Seq Scan on foo (cost=0.00..145.00 rows=10000 width=8)

(actual time=0.021..9.908 rows=10000 loops=1)-> Hash (cost=1.90..1.90 rows=90 width=4)

(actual time=0.280..0.280 rows=90 loops=1)Buckets: 1024 Batches: 1 Memory Usage: 4kB-> Seq Scan on baz (cost=0.00..1.90 rows=90 width=4)

(actual time=0.010..0.138 rows=90 loops=1)

-> Hash (cost=4.00..4.00 rows=100 width=106)(actual time=0.354..0.354 rows=100 loops=1)

Buckets: 1024 Batches: 1 Memory Usage: 14kB-> Seq Scan on bar (cost=0.00..4.00 rows=100 width=106)

(actual time=0.007..0.167 rows=100 loops=1)Total runtime: 59.376 ms


23/32

Not The Same Thing!

SELECT * FROM(foo JOIN bar ON foo.x = bar.x)LEFT JOIN baz ON foo.y = baz.y

SELECT * FROM(foo LEFT JOIN baz ON foo.y = baz.y)JOIN bar ON foo.x = bar.x


24/32

Review of Join Planning

Join Order

Join Strategy

Nested loop Nested loop with inner index-scan

Merge join

Hash join Join removal

Inner vs. outer


25/32

Aggregates and DISTINCT

Plain aggregate.

e.g. SELECT count(*) FROM foo;

Sorted aggregate. Sort the data (or use pre-sorted data); when you

see a new value, aggregate the prior group.

Hashed aggregate.

Insert each input row into a hash table based onthe grouping columns; at the end, aggregate allthe groups.


26/32

Statistics

All of the decisions discussed earlier in this talkare made using statistics.

Seq scan vs. index scan vs. bitmap index scan

Nested loop vs. merge join vs. hash join

ANALYZE (manual or via autovacuum) gathersthis information.

You must have good statistics or you will get badplans!


27/32

Confusing The Planner

SELECT * FROM foo WHERE a = 1 AND b = 1

If 20% of the rows have a = 1 and 10% of the

rows have b = 1, the planner will assume that20% * 10% = 2% of the rows meet both criteria.

SELECT * FROM foo WHERE (a + 0) = a

Planner doesn't have a clue, so will assume 0.5%of rows will match.


28/32

What Could Go Wrong?

If the plannerunderestimates the row count, itmay choose an index scan instead of a sequentialscan, or a nested loop instead of a hash or merge

join.

If the planneroverestimates the row count, it maychoose a sequential scan instead of an index scan,

or a merge or hash join instead of a nested loop. Small values for LIMIT tilt the planner toward

fast-start plans and magnify the effect of badestimates.


29/32

Query Planner Parameters seq_page_cost (1.0), random_page_cost (4.0) Reduce

these costs to account for caching effects. If database isfully cached, try 0.005.

default_statistics_target (10 or 100) Level of detail forstatistics gathering. Can also be overridden on a per-column basis.

enable_hashjoin, enable_sort, etc. - Just for testing.

work_mem Amount of memory per sort or hash. from_collapse_limit, join_collapse_limit,

geqo_threshold Sometimes need to be raised, but becareful!


30/32

Things That Are Slow

DISTINCT.

PL/pgsql loops.FOR x IN SELECT ... LOOP SELECT ... END LOOP

Repeated calls to SQL or PL/pgsql functions.SELECT id, some_function(id) FROM table;


31/32

Upcoming Features

Join removal (right now just for LEFT joins).

Machine-readable EXPLAIN output.

Hash statistics. Better model for Materialize costs.

Improved use of indices to handle MIN(x),

MAX(x), and x IS NOT NULL.


32/32

Questions?

Any questions?

the postgresql query planner_longer

Documents