更新列的值

Question

我有下表：

SELECT * FROM labels LIMIT 5;
 user_id | session_id |    start_timestamp     |     end_timestamp      | travelmode 
---------+------------+------------------------+------------------------+------------
      11 |          0 | 2007-06-26 12:32:29+01 | 2007-06-26 12:40:29+01 | bus
      11 |          0 | 2008-03-31 17:00:08+01 | 2008-03-31 17:09:01+01 | taxi
      11 |          0 | 2008-04-01 01:48:32+01 | 2008-04-01 01:59:23+01 | taxi
      11 |          0 | 2008-04-01 02:00:22+01 | 2008-04-01 02:08:13+01 | walk
      11 |          0 | 2008-04-01 12:22:47+01 | 2008-04-01 12:28:39+01 | taxi
(5 rows)

SELECT * FROM trajectories LIMIT 5;
 user_id |    session_id     |       timestamp        |    lat    |    lon     | alt  
---------+-------------------+------------------------+-----------+------------+------
      11 | 10020080330004134 | 2008-03-30 00:41:34+00 | 36.032647 | 103.850612 | -777
      11 | 10020080330004134 | 2008-03-30 00:42:05+00 | 36.031563 | 103.851273 | -777
      11 | 10020080330004134 | 2008-03-30 00:43:04+00 | 36.028623 | 103.853238 | -777
      11 | 10020080330004134 | 2008-03-30 00:44:03+00 | 36.027323 | 103.854475 | -777
      11 | 10020080330004134 | 2008-03-30 00:45:02+00 | 36.025775 | 103.854993 | -777
(5 rows)

所以我想更新session_id表的labels列（最初全为零）：

UPDATE labels
SET session_id=trajectories.session_id
FROM trajectories
WHERE 
  trajectories.user_id = labels.user_id
  AND trajectories.timestamp >= labels.start_timestamp 
  AND trajectories.timestamp <= labels.end_timestamp;

UPDATE 4500

但是，在执行查询大约5分钟后，并没有更新labels表的所有列（仅约30％）：

 SELECT COUNT(*) FROM labels WHERE session_id=0;
 count 
-------
 10217
(1 row)

如果可能有帮助，请进一步了解表格：

\d labels
                             Table "akil.labels"
     Column      |           Type           | Collation | Nullable | Default 
-----------------+--------------------------+-----------+----------+---------
 user_id         | integer                  |           | not null | 
 session_id      | bigint                   |           |          | 
 start_timestamp | timestamp with time zone |           | not null | 
 end_timestamp   | timestamp with time zone |           | not null | 
 travelmode      | text                     |           |          | 
Indexes:
    "mode_pkey" PRIMARY KEY, btree (user_id, start_timestamp, end_timestamp)

 \d trajectories 
                       Table "akil.trajectories"
   Column   |           Type           | Collation | Nullable | Default 
------------+--------------------------+-----------+----------+---------
 user_id    | integer                  |           |          | 
 session_id | bigint                   |           | not null | 
 timestamp  | timestamp with time zone |           | not null | 
 lat        | double precision         |           | not null | 
 lon        | double precision         |           | not null | 
 alt        | double precision         |           |          | 
Indexes:
    "trajectory_pkey" PRIMARY KEY, btree (session_id, "timestamp", lat, lon)

编辑

向trajectories表添加索引：

 CREATE INDEX idx_trecj ON trajectories (user_id, timestamp);
CREATE INDEX

UPDATE labels
SET session_id=trajectories.session_id
FROM trajectories
WHERE 
  trajectories.user_id = labels.user_id
  AND trajectories.timestamp >= labels.start_timestamp 
  AND trajectories.timestamp <= labels.end_timestamp;

UPDATE 4500

 SELECT COUNT(*) FROM labels WHERE session_id=0;
 count 
-------
 10217
(1 row

但是，并不是session_id中的所有labels都被更新（与初始操作相同）。

Answer 1

对于trajectories，您需要在(user_id, timestamp)上建立索引。

这应该有助于您的表现。

您还存在可能存在多个匹配项的问题。在这种情况下，您当前的update必须做的工作比必要的多。但是索引应该是一个很好的第一步。

Answer 2

search_path

您的表定义重新显示：

表“ akil.labels”

因此表位于模式akil中。是否始终正确设置search_path，以便您有能力在查询中忽略模式限定？否则，您可能是偶然/从错误的表中更新了错误的表-这可以解释您的结果。参见：

• How does the search_path influence identifier resolution and the "current schema"

但是还有其他可能的解释。

akil.trajectories中没有匹配项>>

是什么使您认为trajectories中的每一行都有适用的行？

很多人都没有：

labels
SELECT count(*) AS no_match_in_trajectories
FROM   akil.labels l
WHERE  NOT EXISTS (
   SELECT FROM akil.trajectories t
   WHERE  t.user_id   =  l.user_id
   AND    t.timestamp >= l.start_timestamp 
   AND    t.timestamp <= l.end_timestamp
   );
中有多个匹配项>
有关说明：是什么让您认为会有确切地是

适用行？如果akil.trajectories中有多个适用行，则UPDATE会产生任意结果（以昂贵的方式）。

鉴于您的表定义，这将是正确的方法：

trajectories
关于可避免的空白更新：

• UPDATE akil.labels l
  SET    session_id = t.session_id
  FROM   akil.labels l1
  CROSS  JOIN   LATERAL (
     SELECT t1.session_id
     FROM   akil.trajectories t1
     WHERE  t1.user_id   =  l1.user_id
     AND    t1.timestamp >= l1.start_timestamp 
     AND    t1.timestamp <= l1.end_timestamp
     ORDER  BY t.timestamp  -- pick the earliest matching entry
     LIMIT  1
     ) t
  WHERE ( l.user_id,  l.start_timestamp,  l.end_timestamp)  -- PK
     =  (l1.user_id, l1.start_timestamp, l1.end_timestamp)
  AND    l.session_id IS DISTINCT FROM t.session_id;        -- avoids empty updates

  完美的索引是：

  How do I (or can I) SELECT DISTINCT on multiple columns?

[CREATE INDEX foo ON akil.trajectories (user_id, timestamp) INCLUDE (session_id) 需要Postgres 11或更高版本。

Asides

您的整个设置看起来可疑地凌乱。

• INCLUDE是INCLUDE PK列的名称。颇具误导性，表似乎已重命名？还是那里有东西？

• mode_pkey的PK：

  labels
  同样，非默认名称。首先，跨越4列的PK是可疑的。并且trajectories和"trajectory_pkey" PRIMARY KEY, btree (session_id, "timestamp", lat, lon)
  是浮点数。这会导致重复操作出现舍入错误，从而导致各种混乱。

• 就像我评论过的：如果lat为lon，则该列可能应该为session_id。