选择包含相同值SQL的连续行的最新记录

Question

我有一个程序每小时对一组机器运行一次 ping，并将该 ping 的结果（无论该计算机是否连接）存储在一个名为 machine_ping_data 的表中。

This is the structure of the table

我正在尝试获取status 成功“SUCC”的每台机器的最新连续记录。连续我的意思是他们可以被另一个 machine_id 的记录打断，但不能被同一 machine_id 的记录和不同的 status （UNK，FAIL）打断。 另外，当我说最新时，我的意思是对于machine_id，除了“SUCC”之外，没有更多最近的记录，status。我希望这很清楚。

This is an example of the data I have in the table

我目前的代码如下：

    SELECT machine_id, status, TIMESTAMPDIFF(hour, MIN(insert_date), MAX(insert_date)) AS durationFROM (
        SELECT
        *
        , @groupNumber := IF(@prev_machine_id != machine_id OR @prev_status != status, @groupNumber + 1, @groupNumber) AS gn
        , @prev_machine_id  := machine_id
        , @prev_status := status
        FROM machine_ping_data
        , (SELECT @groupNumber := 0, @prev_machine_id  := NULL, @prev_status := NULL) var_init_subquery
        ORDER BY machine_id, insert_date DESC
    ) sq
    WHERE status = "SUCC"
    GROUP BY gn, machine_id, status

它确实对连续的结果进行分组，但不仅仅是最新的结果，它会获取特定machine_id 的最新记录，而不管status 是否为“SUCC”。 我也尝试过使用联合，但没有得到任何结果。

任何帮助或建议将不胜感激！

Answer 1

首先将您的问题归结为核心：选择最近连续成功的 ping。

这可以通过仅使用包含三列的数据表来完成：(machine_id INT, insert_date INT, status ENUM("SUCC", "FAIL"))。

针对这个问题，我找到了以下解决方案：

SELECT C.machine_id, C.insert_date
  FROM data AS C
  WHERE C.status = "SUCC"
  AND (
    C.insert_date > ANY ( -- C.insert_date is later than the latest failure which occurred before the latest success.
      SELECT max(A.insert_date) AS LATEST_FAILURE_BEFORE_LATEST_SUCCESS
          FROM data AS A JOIN data AS B
          ON (A.machine_id = B.machine_id AND A.status != "SUCC" AND B.status = "SUCC" AND A.insert_date < B.insert_date)
          GROUP BY A.machine_id HAVING A.machine_id = C.machine_id
      )
    OR NOT EXISTS ( -- corner case NO_PRECEDING_FAILURES
      SELECT 1 FROM data AS D 
        WHERE D.machine_id = C.machine_id
        AND D.status != "SUCC"
    )
  );

我想复杂的部分是内部的SELECT，我在其中加入数据并选择LATEST_FAILURE_BEFORE_LATEST_SUCCESS。在这里，我利用成功 ping 的最新连续记录（通常）在它们之前有失败的事实。我选择最新的fa

带有测试用例的完整脚本：

DROP DATABASE example;
CREATE DATABASE example;
USE example;

CREATE TABLE data (
    machine_id INT,
    insert_date INT,
    status ENUM("SUCC", "FAIL")
);

-- Test cases with different machines
INSERT INTO data (machine_id, insert_date, status) VALUES
-- 0. a machine which has sequences of FAIL - SUCC
(0, 4, "FAIL"),
(0, 3, "SUCC"), -- TARGET
(0, 2, "SUCC"), -- TARGET
(0, 1, "FAIL"),
(0, 0, "FAIL"),
-- 1. a machine which has sequences of FAIL - SUCC - FAIL
(1, 4, "FAIL"),
(1, 3, "SUCC"), -- TARGET
(1, 2, "SUCC"), -- TARGET
(1, 1, "FAIL"),
(1, 0, "FAIL"),
-- 2. a machine which has sequences of FAIL - SUCC - FAIL - SUCC
(2, 7, "SUCC"), -- TARGET
(2, 6, "SUCC"), -- TARGET
(2, 5, "SUCC"), -- TARGET
(2, 4, "FAIL"),
(2, 3, "SUCC"),
(2, 2, "SUCC"),
(2, 1, "FAIL"),
(2, 0, "FAIL"),
-- 3. a machine which has sequences of FAIL - SUCC - FAIL - SUCC - FAIL
(3, 9, "FAIL"),
(3, 8, "FAIL"),
(3, 7, "SUCC"), -- TARGET
(3, 6, "SUCC"), -- TARGET
(3, 5, "SUCC"), -- TARGET
(3, 4, "FAIL"),
(3, 3, "SUCC"),
(3, 2, "SUCC"),
(3, 1, "FAIL"),
(3, 0, "FAIL"),
-- 4. a machine which has sequences of SUCC - FAIL
(4, 2, "FAIL"),
(4, 1, "SUCC"), -- TARGET
(4, 0, "SUCC"), -- TARGET
-- 5. a machine which has sequences of SUCC - FAIL - SUCC
(5, 5, "SUCC"), -- TARGET
(5, 4, "SUCC"), -- TARGET
(5, 3, "SUCC"), -- TARGET
(5, 2, "FAIL"),
(5, 1, "SUCC"),
(5, 0, "SUCC"),
-- 6. a machine which has sequences of SUCC - FAIL - SUCC - FAIL
(6, 7, "FAIL"),
(6, 6, "FAIL"),
(6, 5, "SUCC"), -- TARGET
(6, 4, "SUCC"), -- TARGET
(6, 3, "SUCC"), -- TARGET
(6, 2, "FAIL"),
(6, 1, "SUCC"),
(6, 0, "SUCC"),
-- 7. a machine which has only FAIL
(7, 2, "FAIL"),
(7, 1, "FAIL"),
(7, 0, "FAIL"),
-- 8. a machine which has only SUCC
(8, 2, "SUCC"), -- TARGET
(8, 1, "SUCC"), -- TARGET
(8, 0, "SUCC"); -- TARGET

SELECT C.machine_id, C.insert_date
  FROM data AS C
  WHERE C.status = "SUCC"
  AND (
    C.insert_date > ANY ( -- C.insert_date is later than the latest failure which occurred before the latest success.
      SELECT max(A.insert_date) AS LATEST_FAILURE_BEFORE_LATEST_SUCCESS
          FROM data AS A JOIN data AS B
          ON (A.machine_id = B.machine_id AND A.status != "SUCC" AND B.status = "SUCC" AND A.insert_date < B.insert_date)
          GROUP BY A.machine_id HAVING A.machine_id = C.machine_id
      )
    OR NOT EXISTS ( -- corner case NO_PRECEDING_FAILURES
      SELECT 1 FROM data AS D 
        WHERE D.machine_id = C.machine_id
        AND D.status != "SUCC"
    )
  );

-- Should return:
-- machine_id   insert_date
-- 0    3
-- 0    2
-- 1    3
-- 1    2
-- 2    7
-- 2    6
-- 2    5
-- 3    7
-- 3    6
-- 3    5
-- 5    5
-- 5    4
-- 5    3
-- 6    5
-- 6    4
-- 6    3
-- 8    2
-- 8    1
-- 8    0

DROP DATABASE example;

Answer 2

我认为鉴于您要实现的目标的复杂性，最好保持 sql 语句简单：

选择insert_date, machine_id, status FROM machine_ping_data ORDER BY insert_date DESC

然后使用 PHP/无论您调用什么 sql 查询来处理数据

因此，假设 PHP 下面的这段代码可以让您在每台机器上获得每个成功 ping 的日期，直到最近一次不成功。

$sql = "SELECT `insert_date`, `machine_id`, `status` FROM `machine_ping_data` ORDER BY `insert_date` DESC";
$db = (however you establish connection to your sql database...)
$res = $db->query($sql);
$machines = array();
while($r = $res->fetch_object())
{
  if(!isset($machines[$r->machine_id]))
  {
    $x = new stdclass;
    $x->history = array();
    $machines[$r->machine_id] = $x; 
  }

  $x->history[] = $r;
  if($r->status!="SUCC")
  {
    continue;
  }
}

那么每台机器的“历史”数组中的第一个条目是你最近的成功，最后一个条目是你最近失败的那台机器。

根据我的经验，最好让 sql 查询尽可能简单，然后再处理数据；这也减少了服务器的负载。