Select MAX or Order By Limit 1

后端 未结 3 1653
生来不讨喜
生来不讨喜 2021-02-14 03:44

MIN/MAX vs ORDER BY and LIMIT

To follow up on this question: I found some results very different from what Sean McSomething describes:

I have a table with about

相关标签:
3条回答
  • 2021-02-14 04:20

    I've a similar situation, index on the column in question, and yet the order by & limit solution seems quicker. How good is that :)

    0 讨论(0)
  • 2021-02-14 04:36

    To avoid a full pass, add an INDEX on foo column.

    0 讨论(0)
  • 2021-02-14 04:38

    I came across this question and thought I'd add what I've found. Note that the columns are indexed. I'm running on MariaDB 10.2.14.

    I have a query which looks like SELECT MAX(created) FROM tbl WHERE group=0 AND created IS NOT NULL. There's an index on (group,created) (both are ints, but created can be NULL). There are many entries with group=0, not many where created IS NULL. tbl is using the Aria storage engine.

    EXPLAIN shows the index is being used and gives a row count of 46312, with extra saying "Using where; Using index"

    Running the query takes around 0.692s, but the status has something interesting:

    Handler_read_key: 1 Handler_read_next: 45131 Handler_read_prev: 0

    This seems to suggest that the key is being fully scanned for the maximum; using MIN instead of MAX seems to give similar results. This seems to suggest that MIN/MAX actually can't make use of the optimisation to just pick the first/last entry of the index here.

    However, if the query is changed to SELECT created FROM tbl WHERE group=0 AND created IS NOT NULL ORDER BY created DESC LIMIT 1, whilst the query seems to take about the same amount of time to run, and EXPLAIN shows the same info, the status shows:

    Handler_read_key: 1 Handler_read_next: 0 Handler_read_prev: 0

    I get similar results if the order by is changed to ASC. It seems to me that using an ORDER BY...LIMIT can skip an index scan, which could lead to faster queries if there are many rows which match the index condition, if my understanding is correct.
    Note that for the above results, there's enough RAM and cache allocated for holding all indexes in cache, so, presumably, index scans are fast.

    I haven't done any experiments with other conditions (different MySQL versions and storage engines) but I suppose the moral of this story is, checking status of queries via SHOW SESSION STATUS may help provide answers to these things.
    At least in this case, the ORDER BY...LIMIT may be more efficient than MIN/MAX even when an index can be used.

    0 讨论(0)
提交回复
热议问题