diff --git a/choose-index.md b/choose-index.md index b00824139bba4..b2ebd30f61146 100644 --- a/choose-index.md +++ b/choose-index.md @@ -29,21 +29,85 @@ Before introducing index selection, it is important to understand the ways TiDB ## Index selection rules -TiDB provides a heuristic rule named skyline-pruning based on the cost estimation of each operator for accessing tables. It can reduce the probability of wrong index selection caused by wrong estimation. +TiDB selects indexes based on rules or cost. The based rules include pre-rules and skyline-pruning. When selecting an index, TiDB tries the pre-rule first. If an index satisfies a pre-rule, TiDB directly selects this index. Otherwise, TiDB uses skyline-pruning to exclude unsuitable indexes, and then selects the index with the lowest cost based on the cost estimation of each operator that accesses tables. -### Skyline-pruning +### Rule-based selection -Skyline-pruning is a heuristic filtering rule for indexes. To judge an index, the following three dimensions are needed: +#### Pre-rules -- Whether it needs to retrieve rows from a table when you select the index to access the table (that is, the plan generated by the index is IndexReader operator or IndexLookupReader operator). Indexes that do not retrieve rows from a table are better on this dimension than indexes that do. +TiDB uses the following heuristic pre-rules to select indexes: -- Select whether the index satisfies a certain order. Because index reading can guarantee the order of certain column sets, indexes that satisfy the query order are superior to indexes that do not satisfy on this dimension. ++ Rule 1: If an index satisfies "unique index with full match + no need to retrieve rows from a table (which means that the plan generated by the index is the IndexReader operator)", TiDB directly selects this index. + ++ Rule 2: If an index satisfies "unique index + the need to retrieve rows from a table (which means that the plan generated by the index is the IndexReader operator)", TiDB selects the index with the smallest number of rows to be retrieved from a table as a candidate index. + ++ Rule 3: If an index satisfies "ordinary index + no need to retrieve rows from a table + the number of rows to be read is less than the value of a certain threshold", TiDB selects the index with the smallest number of rows to be read as a candidate index. + ++ Rule 4: If only one candidate index is selected based on rule 2 and 3, select this candidate index. If two candidate indexes are respectively selected based on rule 2 and 3, select the index with the smaller number of rows to be read (the number of rows with index + the number of rows to be retrieved from a table). + +The "index with full match" in the above rules means each indexed column has the equal condition. When executing the `EXPLAIN FORMAT = 'verbose' ...` statement, if the pre-rules match an index, TiDB outputs a NOTE-level warning indicating that the index matches the pre-rule. + +In the following example, because the index `idx_b` meets the condition "unique index + the need to retrieve rows from a table" in rule 2, TiDB selects the index `idx_b` as the access path, and `SHOW WARNING` returns a note indicating that the index `idx_b` matches the pre-rule. + +```sql +mysql> CREATE TABLE t(a INT PRIMARY KEY, b INT, c INT, UNIQUE INDEX idx_b(b)); +Query OK, 0 rows affected (0.01 sec) + +mysql> EXPLAIN FORMAT = 'verbose' SELECT b, c FROM t WHERE b = 3 OR b = 6; ++-------------------+---------+---------+------+-------------------------+------------------------------+ +| id | estRows | estCost | task | access object | operator info | ++-------------------+---------+---------+------+-------------------------+------------------------------+ +| Batch_Point_Get_5 | 2.00 | 8.80 | root | table:t, index:idx_b(b) | keep order:false, desc:false | ++-------------------+---------+---------+------+-------------------------+------------------------------+ +1 row in set, 1 warning (0.00 sec) + +mysql> SHOW WARNINGS; ++-------+------+-------------------------------------------------------------------------------------------+ +| Level | Code | Message | ++-------+------+-------------------------------------------------------------------------------------------+ +| Note | 1105 | unique index idx_b of t is selected since the path only has point ranges with double scan | ++-------+------+-------------------------------------------------------------------------------------------+ +1 row in set (0.00 sec) +``` + +### Skyline-pruning + +Skyline-pruning is a heuristic filtering rule for indexes, which can reduce the probability of wrong index selection caused by wrong estimation. To judge an index, the following three dimensions are needed: - How many access conditions are covered by the indexed columns. An “access condition” is a where condition that can be converted to a column range. And the more access conditions an indexed column set covers, the better it is in this dimension. -For these three dimensions, if an index named idx_a is not worse than the index named idx_b in all three dimensions and one of the dimensions is better than idx_b, then idx_a is preferred. +- Whether it needs to retrieve rows from a table when you select the index to access the table (that is, the plan generated by the index is IndexReader operator or IndexLookupReader operator). Indexes that do not retrieve rows from a table are better on this dimension than indexes that do. If both indexes need TiDB to retrieve rows from the table, compare how many filtering conditions are covered by the indexed columns. Filtering conditions mean the `where` condition that can be judged based on the index. If the column set of an index covers more access conditions, the smaller the number of retrieved rows from a table, and the better the index is in this dimension. + +- Select whether the index satisfies a certain order. Because index reading can guarantee the order of certain column sets, indexes that satisfy the query order are superior to indexes that do not satisfy on this dimension. -### Selection based on cost estimation +For these three dimensions above, if the index `idx_a` performs no worse than the index `idx_b` in all three dimensions and performs better than `idx_b` in one dimension, then `idx_a` is preferred. When executing the `EXPLAIN FORMAT = 'verbose' ...` statement, if skyline-pruning excludes some indexes, TiDB outputs a NOTE-level warning listing the remaining indexes after the skyline-pruning exclusion. + +In the following example, the indexes `idx_b` and `idx_e` are both inferior to `idx_b_c`, so they are excluded by skyline-pruning. The returned result of `SHOW WARNING` displays the remaining indexes after skyline-pruning. + +```sql +mysql> CREATE TABLE t(a INT PRIMARY KEY, b INT, c INT, d INT, e INT, INDEX idx_b(b), INDEX idx_b_c(b, c), INDEX idx_e(e)); +Query OK, 0 rows affected (0.01 sec) + +mysql> EXPLAIN FORMAT = 'verbose' SELECT * FROM t WHERE b = 2 AND c > 4; ++-------------------------------+---------+---------+-----------+------------------------------+----------------------------------------------------+ +| id | estRows | estCost | task | access object | operator info | ++-------------------------------+---------+---------+-----------+------------------------------+----------------------------------------------------+ +| IndexLookUp_10 | 33.33 | 738.29 | root | | | +| ├─IndexRangeScan_8(Build) | 33.33 | 2370.00 | cop[tikv] | table:t, index:idx_b_c(b, c) | range:(2 4,2 +inf], keep order:false, stats:pseudo | +| └─TableRowIDScan_9(Probe) | 33.33 | 2370.00 | cop[tikv] | table:t | keep order:false, stats:pseudo | ++-------------------------------+---------+---------+-----------+------------------------------+----------------------------------------------------+ +3 rows in set, 1 warning (0.00 sec) + +mysql> SHOW WARNINGS; ++-------+------+------------------------------------------------------------------------------------------+ +| Level | Code | Message | ++-------+------+------------------------------------------------------------------------------------------+ +| Note | 1105 | [t,idx_b_c] remain after pruning paths for t given Prop{SortItems: [], TaskTp: rootTask} | ++-------+------+------------------------------------------------------------------------------------------+ +1 row in set (0.00 sec) +``` + +### Cost estimation-based selection After using the skyline-pruning rule to rule out inappropriate indexes, the selection of indexes is based entirely on the cost estimation. The cost estimation of accessing tables requires the following considerations: