Partial indexes allow you to specify a subset of rows and columns to add to an index. Partial indexes include the subset of rows in a table that evaluate to true on a boolean predicate expression (i.e., a WHERE filter) defined at index creation.
How do partial indexes work?
When you create a partial index, CockroachDB "indexes" the columns and rows that evaluate to true on the index's boolean predicate expression, creating a sorted copy of the subset of row values, without modifying the values in the table itself.
CockroachDB can use a partial index to efficiently execute queries on any subset of rows implied by the partial index. When possible, the cost-based optimizer creates a plan that limits table scans on rows implied by the partial index to just the rows in the index. It also limits index rewrites to fewer rows.
Partial indexes can improve cluster performance in a number of ways:
- They contain fewer rows than full indexes, making them less expensive to create and store on a cluster.
- Read queries on rows included in a partial index only scan the rows in the partial index. This contrasts with queries on columns in full indexes, which must scan all rows in the indexed column.
- Write queries on tables with a partial index only perform an index write when the rows inserted satisfy the partial index predicate. This contrasts with write queries on tables with full indexes, which incur the overhead of a full index write when the rows inserted modify an indexed column.
When a query on a table with a partial index has a filter expression, the cost-based optimizer attempts to prove that the filter implies the partial index predicate. It is not guaranteed that the optimizer can prove the implication of arbitrarily complex expressions. Although unlikely, it is possible that a filter implies a predicate, but the optimizer cannot prove the implication.
Creation
To create a partial index, use a CREATE INDEX statement, with a standard WHERE clause defining a predicate expression.
For example, to define a partial index on columns a and b of table t, filtering on rows in column c greater than 5:
> CREATE INDEX ON t (a, b) WHERE c > 5;
The following queries use the partial index:
> SELECT a, b FROM t WHERE c > 5;
> SELECT * FROM t WHERE c = 10;
The following queries do not use the partial index:
> SELECT a, b FROM t;
> SELECT * FROM t WHERE c = 3;
When defining the predicate expression, note that:
- The predicate expression must result in a boolean.
- The predicate expression can only refer to columns in the table being indexed.
- Functions used in predicates must be immutable. For example, the
now()function is not allowed in predicates because its value depends on more than its arguments.
Unique partial indexes
You can enforce uniqueness on a subset of rows with CREATE UNIQUE INDEX ... WHERE ....
For example, to define a unique partial index on columns a and b for table t, filtering on rows in column d equal to 'x':
> CREATE UNIQUE INDEX ON t (a, b) WHERE d = 'x';
This creates a partial index and a UNIQUE constraint on the subset of rows where d is equal to 'x'.
For another example, see Create a partial index that enforces uniqueness on a subset of rows.
When inserted values conflict with a UNIQUE constraint on one or more columns, CockroachDB normally returns an error. We recommend adding an ON CONFLICT clause to all INSERT statements that might conflict with rows in the unique index.
Partial GIN indexes
You can create partial GIN indexes, which are indexes on a subset of JSON, ARRAY, or geospatial container column data.
Index hints
You can force queries to use a specific partial index (also known as "index hinting"), like you can with full indexes. However, unlike full indexes, partial indexes cannot be used to satisfy all queries. If a query's filter implies the partial index predicate expression, the partial index will be used in the query plan. If not, an error will be returned.
Examples
Setup
The following examples use the movr example dataset.
Start the MovR database on a 3-node CockroachDB demo cluster with a larger data set.
cockroach demo movr --num-histories 250000 --num-promo-codes 250000 --num-rides 125000 --num-users 12500 --num-vehicles 3750 --nodes 3
Create an index on a subset of rows
Suppose that you want to query the subset of rides with a revenue greater than 90.
> WITH x AS (SHOW TABLES) SELECT * FROM x WHERE table_name='rides';
schema_name | table_name | type | owner | estimated_row_count | locality
--------------+------------+-------+-------+---------------------+-----------
public | rides | table | demo | 125000 | NULL
(1 row)
Time: 21ms total (execution 21ms / network 0ms)
Without a partial index, querying the rides table with a WHERE revenue > 90 clause will scan the entire table. To see the plan for such a query, you can use an EXPLAIN statement:
> EXPLAIN SELECT * FROM rides WHERE revenue > 90;
info
------------------------------------------------------------------------------------------
distribution: full
vectorized: true
• filter
│ estimated row count: 12,472
│ filter: revenue > 90
│
└── • scan
estimated row count: 125,000 (100% of the table; stats collected 12 seconds ago)
table: rides@rides_pkey
spans: FULL SCAN
(11 rows)
Time: 1ms total (execution 1ms / network 0ms)
The estimated row count in the scan node lists the number of rows that the query plan will scan (in this case, the entire table row count of 125,000). The table property lists the index used in the scan (in this case, the primary key index).
To limit the number of rows scanned to just the rows that you are querying, you can create a partial index:
> CREATE INDEX ON rides (city, revenue) WHERE revenue > 90;
> SHOW INDEXES FROM rides;
table_name | index_name | non_unique | seq_in_index | column_name | direction | storing | implicit | visible
-------------+-----------------------------------------------+------------+--------------+---------------+-----------+---------+----------+----------
rides | rides_auto_index_fk_city_ref_users | t | 1 | city | ASC | f | f | t
rides | rides_auto_index_fk_city_ref_users | t | 2 | rider_id | ASC | f | f | t
rides | rides_auto_index_fk_city_ref_users | t | 3 | id | ASC | f | t | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 1 | vehicle_city | ASC | f | f | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 2 | vehicle_id | ASC | f | f | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 3 | city | ASC | f | t | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 4 | id | ASC | f | t | t
rides | rides_city_revenue_idx | t | 1 | city | ASC | f | f | t
rides | rides_city_revenue_idx | t | 2 | revenue | ASC | f | f | t
rides | rides_city_revenue_idx | t | 3 | id | ASC | f | t | t
rides | rides_pkey | f | 1 | city | ASC | f | f | t
rides | rides_pkey | f | 2 | id | ASC | f | f | t
rides | rides_pkey | f | 3 | vehicle_city | N/A | t | f | t
rides | rides_pkey | f | 4 | rider_id | N/A | t | f | t
rides | rides_pkey | f | 5 | vehicle_id | N/A | t | f | t
rides | rides_pkey | f | 6 | start_address | N/A | t | f | t
rides | rides_pkey | f | 7 | end_address | N/A | t | f | t
rides | rides_pkey | f | 8 | start_time | N/A | t | f | t
rides | rides_pkey | f | 9 | end_time | N/A | t | f | t
rides | rides_pkey | f | 10 | revenue | N/A | t | f | t
(20 rows)
Time: 15ms total (execution 14ms / network 1ms)
Another EXPLAIN statement shows that the number of rows scanned by the original query decreases significantly with a partial index on the rides table:
> EXPLAIN SELECT * FROM rides WHERE revenue > 90;
info
------------------------------------------------------------------------------------------
distribution: full
vectorized: true
• index join
│ estimated row count: 12,472
│ table: rides@rides_pkey
│
└── • scan
estimated row count: 12,472 (10.0% of the table; stats collected 36 seconds ago)
table: rides@rides_city_revenue_idx (partial index)
spans: FULL SCAN
(11 rows)
Time: 1ms total (execution 1ms / network 0ms)
Note that the query's SELECT statement queries all columns in the rides table, not just the indexed columns. As a result, an "index join" is required on both the primary index and the partial index.
Querying only the columns in the index will make the query more efficient by removing the index join from the query plan:
> EXPLAIN SELECT city, revenue FROM rides WHERE revenue > 90;
info
------------------------------------------------------------------------------------
distribution: full
vectorized: true
• scan
estimated row count: 11,463 (9.2% of the table; stats collected 4 minutes ago)
table: rides@rides_city_revenue_idx (partial index)
spans: FULL SCAN
(7 rows)
Time: 1ms total (execution 1ms / network 0ms)
Querying a subset of the rows implied by the partial index predicate expression (in this case, revenue > 90) will also use the partial index:
> EXPLAIN SELECT city, revenue FROM rides WHERE revenue > 95;
info
----------------------------------------------------------------------------------------
distribution: full
vectorized: true
• filter
│ estimated row count: 5,037
│ filter: revenue > 95
│
└── • scan
estimated row count: 11,463 (9.2% of the table; stats collected 5 minutes ago)
table: rides@rides_city_revenue_idx (partial index)
spans: FULL SCAN
(11 rows)
Time: 1ms total (execution 1ms / network 0ms)
The number of rows scanned is the same, and an additional filter is applied to the query plan so that only the subset specified by the filter is returned.
So far, all the query scans in this example have spanned the entire partial index (i.e., performed a FULL SCAN of the index). This is because the WHERE clause does not filter on the first column in the index prefix (city). Filtering the query on both columns in the partial index will limit the scan to just the rows that match the filter:
> EXPLAIN SELECT city, revenue FROM rides WHERE city = 'new york' AND revenue > 90;
info
-----------------------------------------------------------------------------------
distribution: local
vectorized: true
• scan
estimated row count: 1,301 (1.0% of the table; stats collected 6 minutes ago)
table: rides@rides_city_revenue_idx (partial index)
spans: [/'new york' - /'new york']
(7 rows)
Time: 1ms total (execution 1ms / network 0ms)
Refining the revenue filter expression to match just a subset of the partial index will lower the scanned row count even more:
> EXPLAIN SELECT city, revenue FROM rides WHERE city = 'new york' AND revenue >= 90 AND revenue < 95;
info
---------------------------------------------------------------------------------------
distribution: local
vectorized: true
• filter
│ estimated row count: 746
│ filter: (revenue >= 90) AND (revenue < 95)
│
└── • scan
estimated row count: 14,187 (11% of the table; stats collected 6 minutes ago)
table: rides@rides_pkey
spans: [/'new york' - /'new york']
(11 rows)
Time: 1ms total (execution 1ms / network 0ms)
Create an index that excludes values
Suppose that you have a number of rows in a table with values that you regularly filter out of selection queries (e.g., NULL values).
A selection query on these values will require a full table scan, using the primary index, as shown by the EXPLAIN statement below:
> EXPLAIN SELECT * FROM rides WHERE end_time IS NOT NULL;
info
-----------------------------------------------------------------------------------------
distribution: full
vectorized: true
• filter
│ estimated row count: 125,000
│ filter: end_time IS NOT NULL
│
└── • scan
estimated row count: 125,000 (100% of the table; stats collected 7 minutes ago)
table: rides@rides_pkey
spans: FULL SCAN
(11 rows)
Time: 1ms total (execution 1ms / network 0ms)
You can create a partial index that excludes these rows, making queries that filter out the non-NULL values more efficient.
> CREATE INDEX ON rides (city, revenue) WHERE end_time IS NOT NULL;
> SHOW INDEXES FROM rides;
table_name | index_name | non_unique | seq_in_index | column_name | direction | storing | implicit | visible
-------------+-----------------------------------------------+------------+--------------+---------------+-----------+---------+----------+----------
rides | rides_auto_index_fk_city_ref_users | t | 1 | city | ASC | f | f | t
rides | rides_auto_index_fk_city_ref_users | t | 2 | rider_id | ASC | f | f | t
rides | rides_auto_index_fk_city_ref_users | t | 3 | id | ASC | f | t | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 1 | vehicle_city | ASC | f | f | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 2 | vehicle_id | ASC | f | f | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 3 | city | ASC | f | t | t
rides | rides_auto_index_fk_vehicle_city_ref_vehicles | t | 4 | id | ASC | f | t | t
rides | rides_city_revenue_idx | t | 1 | city | ASC | f | f | t
rides | rides_city_revenue_idx | t | 2 | revenue | ASC | f | f | t
rides | rides_city_revenue_idx | t | 3 | id | ASC | f | t | t
rides | rides_city_revenue_idx1 | t | 1 | city | ASC | f | f | t
rides | rides_city_revenue_idx1 | t | 2 | revenue | ASC | f | f | t
rides | rides_city_revenue_idx1 | t | 3 | id | ASC | f | t | t
rides | rides_pkey | f | 1 | city | ASC | f | f | t
rides | rides_pkey | f | 2 | id | ASC | f | f | t
rides | rides_pkey | f | 3 | vehicle_city | N/A | t | f | t
rides | rides_pkey | f | 4 | rider_id | N/A | t | f | t
rides | rides_pkey | f | 5 | vehicle_id | N/A | t | f | t
rides | rides_pkey | f | 6 | start_address | N/A | t | f | t
rides | rides_pkey | f | 7 | end_address | N/A | t | f | t
rides | rides_pkey | f | 8 | start_time | N/A | t | f | t
rides | rides_pkey | f | 9 | end_time | N/A | t | f | t
rides | rides_pkey | f | 10 | revenue | N/A | t | f | t
(23 rows)
Time: 12ms total (execution 12ms / network 1ms)
> EXPLAIN SELECT (city, revenue) FROM rides WHERE end_time IS NOT NULL;
info
-----------------------------------------------------------------------------------------
distribution: full
vectorized: true
• render
│ estimated row count: 125,000
│
└── • scan
estimated row count: 125,000 (100% of the table; stats collected 8 minutes ago)
table: rides@rides_city_revenue_idx1 (partial index)
spans: FULL SCAN
(10 rows)
Time: 1ms total (execution 1ms / network 0ms)
Create a partial index that enforces uniqueness on a subset of rows
Suppose that you want to constrain a subset of the rows in a table, such that all values for a particular column in the subset are unique. For example, let's say that every user in New York City must have a unique name.
You can do this efficiently with a unique partial index:
> CREATE UNIQUE INDEX ON users (name) WHERE city='new york';
This creates a partial index and a UNIQUE constraint on just the subset of rows where city='new york'.
> SELECT id, name FROM users WHERE city='new york' LIMIT 3;
id | name
---------------------------------------+-------------------
8647a7cf-4af0-4c82-9344-224097f87b1a | Andre Sanchez
598eaab2-5200-40cb-8e19-244d49f3f63a | Austin Meyer
147ae147-ae14-4b00-8000-000000000004 | Catherine Nelson
(3 rows)
> INSERT INTO users(id, city, name) VALUES (gen_random_uuid(), 'new york', 'Andre Sanchez');
ERROR: duplicate key value (name)=('Andre Sanchez') violates unique constraint "users_name_key"
SQLSTATE: 23505
Because the unique partial index predicate only implies the rows where city='new york', the UNIQUE constraint does not apply to all rows in the table.
> INSERT INTO users(id, city, name) VALUES (gen_random_uuid(), 'seattle', 'Andre Sanchez');
INSERT 1
Known limitations
- CockroachDB prevents a column from being dropped using
ALTER TABLE ... DROP COLUMNif it is referenced by a partial index predicate. To drop such a column, the partial indexes need to be dropped first usingDROP INDEX. #97813. - CockroachDB does not currently support multiple arbiter indexes for
INSERT ON CONFLICT DO UPDATE, and will return an error if there are multiple unique or exclusion constraints matching theON CONFLICT DO UPDATEspecification. #53170
See also
Was this helpful?
