Question

今天，我们的生产数据库（Aurora PostgreSQL 9.6.3）遇到了死锁情况，其中多个进程试图在单行上执行相同的UPDATE查询。我们认为死锁只能在无序更新多行时发生，所以这是一个惊喜;然而，它确实发生在我们当天最忙碌的时刻。

这是我们的Python代码中包含UPDATE语句的事务（它是一个穷人的UPSERT）：

with self.connection.cursor() as cursor:
    cursor.execute("""UPDATE students SET name = %s WHERE uuid = %s AND activity_id = %s""", (name, uuid, activityId))

    if cursor.rowcount <= 0:
        cursor.execute("""INSERT INTO students (name, uuid, activity_id) VALUES (%s, %s, %s)""", (name, uuid, activityId))

    if cursor.rowcount <= 0:
        self.connection.rollback()
        raise BaseDao.NotUpserted("No student name was updated or inserted for activity_id %d and uuid %s" % (activityId, uuid))
    else:
        self.connection.commit()

以下是日志中的一些相关行，包括仅更新一行的简单查询：

...
2018-01-19 16:21:27 UTC:[38161]:ERROR:  deadlock detected
2018-01-19 16:21:27 UTC:[38161]:DETAIL:  Process 38161 waits for ShareLock on transaction 90490253; blocked by process 25147.
    Process 25147 waits for ShareLock on transaction 90490267; blocked by process 38161.
    Process 38161: UPDATE students SET name = 'foobar' WHERE uuid = 'ca1b2d153cbdc9574cce' AND activity_id = 35473237
    Process 25147: UPDATE students SET name = 'foobar' WHERE uuid = 'ca1b2d153cbdc9574cce' AND activity_id = 35473237
...

以下是两个相关的表格：

db=> \d students
                                        Table "public.students"
    Column   |          Type          |                             Modifiers                             
-------------+------------------------+-------------------------------------------------------------------
 id          | integer                | not null default nextval('students_id_seq'::regclass)
 name        | character varying(128) | not null
 uuid        | character varying(40)  | not null
 activity_id | integer                | not null
Indexes:
    "students_pkey" PRIMARY KEY, btree (id)
    "students_activity_id" btree (activity_id)
Foreign-key constraints:
    "activity_id_refs_id_76c08098" FOREIGN KEY (activity_id) REFERENCES activities(id) DEFERRABLE INITIALLY DEFERRED

db=> \d activities
                                       Table "public.activities"
      Column       |           Type           |                              Modifiers                               
-------------------+--------------------------+----------------------------------------------------------------------
 id                | integer                  | not null default nextval('activities_id_seq'::regclass)
 start_time        | timestamp with time zone | not null
 end_time          | timestamp with time zone | 
 activity_type     | character varying(2)     | not null
 activity_id       | integer                  | not null
 started_by_id     | integer                  | not null
 activity_state    | integer                  | not null
 legacy_id         | integer                  | 
 hide_report       | boolean                  | not null
 report_status     | integer                  | 
 students_finished | text                     | not null
 room_name         | text                     | 
 last_updated      | timestamp with time zone | 
 state             | integer                  | 
Indexes:
    "activities_pkey" PRIMARY KEY, btree (id)
    "activities_end_time" btree (end_time)
    "activities_room_name_c1f9997a_like" btree (room_name text_pattern_ops)
    "activities_room_name_c1f9997a_uniq" btree (room_name)
    "activities_started_by_id" btree (started_by_id)
Foreign-key constraints:
    "started_by_id_refs_id_5ea35c7a" FOREIGN KEY (started_by_id) REFERENCES users(id) DEFERRABLE INITIALLY DEFERRED
Referenced by:
    TABLE "students" CONSTRAINT "activity_id_refs_id_76c08098" FOREIGN KEY (activity_id) REFERENCES activities(id) DEFERRABLE INITIALLY DEFERRED

当只更新一行时，我们如何陷入这样的死锁？

Answer 1

我可以想到造成这种僵局的两个原因：

执行更新的事务包含多个语句，而其他语句也会创建锁定。
涉及创建额外锁定的触发器。

请记住，死锁不是错误，除非它们经常发生;无法处理死锁是一个错误。只需重试失败的交易。

PostgreSQL死锁仅更新一行

1 个答案: