Thread: Wrong estimate in query plan
In a complex query I have query I noticed that the planner made a bad estimate for a join between two tables in the query (I made sure statistics were up to date).
The join was on a single column. In the first table are about 985 rows. All rows except one have a NULL value in the join column the one value in that column is 1. The other table has 237240 rows and all rows have a 1 in the join column. This column cannot contain NULL values. There is a foreign key constraint between the join columns.
In pgAdmin I had a look at the statistices for the two columns.
Column in first table
Null Fraction 0.998985
Average Width 4
Distinct Values -1
Most Common Values
Most Common Frequencies
Histogram Bounds
Correlation
What I noticed is that are no most common values mentioned ofcourse the value 1 only occurs once in the column but as all other values are NULL you could argue it is a common value.
Column in second table:
Null Fraction 0
Average Width 4
Distinct Values 1
Most Common Values {1}
Most Common Frequencies {1}
Histogram Bounds
Correlation 1
Looks fine :)
Relevant part of AXPLAIN ANALYZE output
' -> Hash Join (cost=40.16..6471.62 rows=241 width=58) (actual time=0.486..102.979 rows=237240 loops=1)'
' Hash Cond: (kb.filiaal_id = fil.filiaal)'
' -> Seq Scan on kassabon kb (cost=0.00..5539.40 rows=237240 width=42) (actual time=0.036..28.562 rows=237240 loops=1)'
' -> Hash (cost=27.85..27.85 rows=985 width=20) (actual time=0.434..0.434 rows=1 loops=1)'
' Buckets: 1024 Batches: 1 Memory Usage: 1kB'
' -> Seq Scan on relatie fil (cost=0.00..27.85 rows=985 width=20) (actual time=0.003..0.382 rows=985 loops=1)'
Notice how it expects 240 rows but gets 237240.
I wondered if this should be reported as a bug? It goes wrong I think because of the statistics of the column in the first table give no information about the values present except the NULL values.
Another thought I had was that the planner could have known there was a matching row in the first table for each row in the second table because there is a foreign key constraint between the two.
Regards, Eelke
Eelke Klein wrote > What I noticed is that are no most common values mentioned ofcourse the > value 1 only occurs once in the column but as all other values are NULL > you > could argue it is a common value. A random sampling is unlikely to choose a record that only appears in 0.1 percent of the table. Two sequential scans plus a hash seems like a good plan. The smaller table is so small a sequential scan is fast. The larger table experts to have all records read so it to should be scanned. Combining with a hash seems sound. The fact the cross-column estimate is way off isn't that big a deal though I'd be curious to hear Tom's opinion on why this is so for educational purposes. David J. -- View this message in context: http://postgresql.1045698.n5.nabble.com/Wrong-estimate-in-query-plan-tp5775727p5775785.html Sent from the PostgreSQL - general mailing list archive at Nabble.com.