FILLFACTOR is probably too big

Back to Catalog
Goal This query identifies base tables with a potentially suboptimal FILLFACTOR setting, targeting those that likely undergo UPDATE operations. A high FILLFACTOR on such tables can lead to poor performance by reducing the likelihood of Heap-Only Tuple (HOT) updates. When a new row version cannot be stored on the same page as the old version, all indexes on the table must be updated with the new tuple's location.
Notes The query assumes that if the FILLFACTOR has been changed to other value than 100 and then back to 100, then the corresponding reloption is still registered. There are more than one possible reloptions. Thus, instead of using the condition reloptions IS NOT NULL, the query filters out reloptions that specify a new FILLFACTOR value. The query considers both column names in English and Estonian.
Type Problem detection (Each row in the result could represent a flaw in the design)
Reliability Medium (Medium number of false-positive results)
License MIT License
Fixing Suggestion Change FILLFACTOR to 90 or 95. It does not have to be smaller because old versions of rows will be soon deleted from blocks by the vacuum process, i.e., it frees space. After changing the FILLFACTOR of an existing table, one has to VACUUM FULL the table to reorganize its internal storage.
Data Source INFORMATION_SCHEMA+system catalog
SQL Query 
WITH base_tables AS (SELECT 
  pg_class.relname AS table_name, 
  pg_namespace.nspname AS table_schema,
 reloptions AS reloptions
FROM 
  pg_catalog.pg_class, 
  pg_catalog.pg_namespace
WHERE 
  pg_class.relnamespace = pg_namespace.oid AND relkind='r'),
  base_tables_reloptions AS (SELECT table_name, table_schema, unnest(reloptions) AS reloptions
 FROM  base_tables),
base_tables_fillfactor AS (SELECT table_schema, table_name, regexp_replace(reloptions,'[^0-9]','','g')::int AS fillfactor
FROM base_tables_reloptions
WHERE reloptions ILIKE 'fillfactor%'),
high_ff AS (SELECT table_schema, table_name
FROM base_tables AS bt
WHERE table_schema NOT IN (SELECT schema_name
FROM INFORMATION_SCHEMA.schemata
WHERE schema_name<>'public' AND
schema_owner='postgres' AND schema_name IS NOT NULL)
AND NOT EXISTS (SELECT 1
FROM base_tables_fillfactor AS btf
WHERE bt.table_schema=btf.table_schema 
AND bt.table_name=btf.table_name
AND btf.fillfactor<>100)),
state_classif AS (SELECT A.table_schema, A.table_name , A.column_name
FROM information_schema.columns A INNER JOIN information_schema.tables T USING (table_schema, table_name)
INNER JOIN information_schema.schemata S
ON A.table_schema=S.schema_name
WHERE T.table_type='BASE TABLE'
AND A.column_name ~*'(seisund|state|olek)'
AND (A.table_schema, A.table_name , A.column_name) NOT IN
(
SELECT 
  table_constraints.table_schema, 
  table_constraints.table_name, 
  constraint_column_usage.column_name
FROM 
  information_schema.table_constraints INNER JOIN 
  information_schema.constraint_column_usage
  USING (table_schema, table_name, constraint_name)
WHERE 
  table_constraints.constraint_type IN ('PRIMARY KEY', 'UNIQUE')
) AND (A.table_schema = 'public'
OR S.schema_owner<>'postgres')),
reasons AS (SELECT table_schema, table_name, 'Contains optional columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c
WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.is_nullable='YES')
UNION SELECT table_schema, table_name, 'Contains boolean columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type = 'boolean')
UNION SELECT table_schema, table_name, 'Contains long textual columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND ((c.data_type = 'text') OR (c.data_type='character varying' AND c.character_maximum_length IS NULL)))
UNION SELECT table_schema, table_name, 'Have state classifier' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM state_classif AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name)
UNION SELECT table_schema, table_name, 'Contains last update time' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type ~*'(timestamp|date)'
AND c.column_name~*'(muutmis|update|change|modify)')
UNION SELECT table_schema, table_name, 'Date or timestamp could be initially missing, i.e., it will be added later' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type ~*'(timestamp|date)'
AND c.column_default~*'infinity')
UNION SELECT table_schema, table_name, 'User-entered code could be changed' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.column_name~*'(kood|code)')
)
SELECT table_schema, table_name, string_agg(reason, ';' ORDER BY reason) AS indications_that_the_table_will_be_updated, 
Count(*) AS the_number_of_different_indications, 
(SELECT Count(*) AS cnt FROM pg_indexes AS pi WHERE pi.schemaname=r.table_schema AND pi.tablename=r.table_name) AS number_of_indexes
FROM reasons AS r
GROUP BY table_schema, table_name
ORDER BY Count(*) DESC, number_of_indexes DESC, table_schema, table_name;

SQL statements that help generate fixes for the identified problem.

SQL Query to Generate FixDescription
WITH base_tables AS (SELECT 
  pg_class.relname AS table_name, 
  pg_namespace.nspname AS table_schema,
 reloptions AS reloptions
FROM 
  pg_catalog.pg_class, 
  pg_catalog.pg_namespace
WHERE 
  pg_class.relnamespace = pg_namespace.oid AND relkind='r'),
  base_tables_reloptions AS (SELECT table_name, table_schema, unnest(reloptions) AS reloptions
 FROM  base_tables),
base_tables_fillfactor AS (SELECT table_schema, table_name, regexp_replace(reloptions,'[^0-9]','','g')::int AS fillfactor
FROM base_tables_reloptions
WHERE reloptions ILIKE 'fillfactor%'),
high_ff AS (SELECT table_schema, table_name
FROM base_tables AS bt
WHERE table_schema NOT IN (SELECT schema_name
FROM INFORMATION_SCHEMA.schemata
WHERE schema_name<>'public' AND
schema_owner='postgres' AND schema_name IS NOT NULL)
AND NOT EXISTS (SELECT 1
FROM base_tables_fillfactor AS btf
WHERE bt.table_schema=btf.table_schema 
AND bt.table_name=btf.table_name
AND btf.fillfactor<>100)),
state_classif AS (SELECT A.table_schema, A.table_name , A.column_name
FROM information_schema.columns A
INNER JOIN information_schema.tables T
ON A.table_schema = T.table_schema
AND A.table_name = T.table_name
INNER JOIN information_schema.schemata S
ON A.table_schema=S.schema_name
WHERE T.table_type='BASE TABLE'
AND A.column_name ~*'(seisund|state|olek)' 
AND (A.table_schema, A.table_name , A.column_name) NOT IN
(
SELECT 
  table_constraints.table_schema, 
  table_constraints.table_name, 
  constraint_column_usage.column_name
FROM 
  information_schema.table_constraints, 
  information_schema.constraint_column_usage
WHERE 
  table_constraints.table_catalog = constraint_column_usage.table_catalog AND
  table_constraints.table_schema = constraint_column_usage.table_schema AND
  table_constraints.table_name = constraint_column_usage.table_name AND
  table_constraints.constraint_name = constraint_column_usage.constraint_name AND
  table_constraints.constraint_type IN ('PRIMARY KEY', 'UNIQUE')
) AND (A.table_schema = 'public'
OR S.schema_owner<>'postgres')),
reasons AS (SELECT table_schema, table_name, 'Contains optional columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.is_nullable='YES')
UNION SELECT table_schema, table_name, 'Contains boolean columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type = 'boolean')
UNION SELECT table_schema, table_name, 'Contains long textual columns' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND ((c.data_type = 'text') OR (c.data_type='character varying' AND c.character_maximum_length IS NULL)))
UNION SELECT table_schema, table_name, 'Have state classifier' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM state_classif AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name)
UNION SELECT table_schema, table_name, 'Contains last update time' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type ~*'(timestamp|date)'
AND c.column_name~*'(muutmis|update|change|modify)')
UNION SELECT table_schema, table_name, 'Date or timestamp could be initially missing, i.e., it will be added later' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.data_type ~*'(timestamp|date)'
AND c.column_default~*'infinity')
UNION SELECT table_schema, table_name, 'User-entered code could be changed' AS reason
FROM high_ff AS f
WHERE EXISTS (SELECT 1
FROM information_schema.columns AS c WHERE c.table_schema=f.table_schema AND c.table_name=f.table_name
AND c.column_name~*'(kood|code)'))
SELECT DISTINCT format('ALTER TABLE %1$I.%2$I SET (FILLFACTOR=90); VACUUM FULL %1$I.%2$I;', table_schema, table_name) AS statements
FROM reasons AS r;
Change the fillfactor to 90.
Collections

This query belongs to the following collections:

NameDescription
Find problems automaticallyQueries, that results point to problems in the database. Each query in the collection produces an initial assessment. However, a human reviewer has the final say as to whether there is a problem or not .
Categories

This query is classified under the following categories:

NameDescription
Data at the database physical levelQueries of this category provide information about the disk usage.
PerformanceQueries of this category provide information about indexes in a database.
Result quality depends on namesQueries of this category use names (for instance, column names) to try to guess the meaning of a database object. Thus, the goodness of names determines the number of false positive and false negative results.