SQL Cursors…Any use cases you would defend?

后端 未结 10 638
萌比男神i
萌比男神i 2021-01-06 23:01

I\'ll go first.

I\'m 100% in the set-operations camp. But what happens when the set logic on the entire desired input domain leads to a such a large retrieval that

相关标签:
10条回答
  • 2021-01-06 23:29

    Along with what David B said, I, too, prefer the loop/table approach.

    With that out of the way, one use case for cursors and the loop/table approach involves extremely large updates. Let's say you have to update 1 billion rows. In many instances, this may not need to be transactional. For example, it might be a data warehouse aggregation where you have the potential to reconstruct from source files if things go south.

    In this case, it may be best to do the update in "chunks", perhaps 1 million or 10 million rows at a time. This helps keep resource usage to a minimum, and allows concurrent use of the machine to be maximized while you update that billion rows. A looped/chunked approach might be best here. Billion row updates on less-than-stellar hardware tend to cause problems.

    0 讨论(0)
  • 2021-01-06 23:29

    Cursors are also handy when you want to run a system proc multiple times with different input values. I have no intention of trying to rewrite system procs to be set-based, so I will use a cursor then. Plus you are usually going through a very limited number of objects. You can do the same thing with an existing proc that inserts only one record at a time, but from a performance view, this is usually a bad thing if you have alot of records to run through. Those I will rewrite to be set-based.

    Running totals as discussed by others can be faster.

    If you are emailing from the database (not the best idea but sometimes it is what you are stuck with), then a cursor can ensure that customer a doesn't see customer b's email address when you send both the same email.

    0 讨论(0)
  • 2021-01-06 23:34

    Very occasionally you will get an operation that needs a cursor but in T-SQL it is fairly rare. Identity(int) columns or sequences order things in ways within set operations. Aggregations where calculations might change at certain points (such as accumulating claims from ground up to a limit or excess point) are inherently procedural, so those are a candidate for a cursor.

    Other candidates would be inherently procedural such as looping through a configuration table and generating and executing a series of queries.

    0 讨论(0)
  • 2021-01-06 23:38

    Sure, there are a number of places where cursors might be better than set-based operations.

    One is if you're updating a lot of data in a table (for example a SQL Agent job to pre-compute data on a schedule) then you might use cursors to do it in multiple small sets rather than one large one to reduce the amount of concurrent locking and thus reduce the chance of lock contention and/or deadlocks with other processes accessing the data.

    Another is if you want to take application-level locks using the sp_getapplock stored procedure, which is useful when you want to ensure rows that are being polled for by multiple processes are retrieved exactly once (example here).

    In general though, I'd agree that it's best to start using set based operations if possible, and only move to cursors if required either for functionality or performance reasons (with evidence to back the latter up).

    0 讨论(0)
  • 2021-01-06 23:39

    In a pure SQL environment, I'd rather avoid cursors as you suggest. But once you cross over into procedural language (like PL/SQL), there are a number of uses. For example, if you want to retrieve certain rows and want "to do" something more complex than update it with them.

    0 讨论(0)
  • 2021-01-06 23:39

    Having to use a cursor is generally a sign that you are doing in the database what ought to be done in the application. As others have said, cursors are generally needed when a stored procedure is calculating running totals, or when you're generating code and/or meta-programming.

    But why are you doing that kind of work in a stored procedure in the first place? Is that really the best use of your database server? Is T-SQL really the right language to use when generating code?

    Sure, sometimes the answer is "yes," or, more likely, "no, but it's simpler this way." In my view, keeping things simple trumps premature optimization any day of the week. So I use cursors. But when I think I need to use a cursor, the universe is asking me a question that I should really have a good answer to.

    0 讨论(0)
提交回复
热议问题