Question

表output_values_center1（和其他一些）继承output_values。我会定期截断表output_values_center1并加载新数据（在一个事务中）。在那个时候用户可以请求一些数据，他收到了错误信息。为什么会发生（选择查询请求只有一条记录）以及如何避免这样的问题：

2010-05-19 14:43:17 UTC ERROR:  deadlock detected
2010-05-19 14:43:17 UTC DETAIL:  Process 25972 waits for AccessShareLock on relation  2495092 of database 16385; blocked by process 26102.
    Process 26102 waits for AccessExclusiveLock on relation 2494865 of database 16385; blocked by process 25972.
    Process 25972: SELECT * FROM "output_values" WHERE ("output_values".id = 122312)  LIMIT 1
    Process 26102: TRUNCATE TABLE "output_values_center1"

Answer 1

"TRUNCATE acquires an ACCESS EXCLUSIVE lock on each table it operates on, which blocks all other concurrent operations on the table. If concurrent access to a table is required, then the DELETE command should be used instead."

显然，目前尚不清楚您是否只查看上面链接的“联机帮助页”，查询父表会影响其后代。 SELECT命令的“联机帮助页”中的以下摘录说明了这一点：

"If ONLY is specified, only that table is scanned. If ONLY is not specified, the table and any descendant tables are scanned."

Answer 2

我会尝试这种（在伪代码中）截断：

#define NOWAIT_TIMES 100
#define SLEEPTIME_USECS (1000*100)
for ( i=0; ; i++ ) {
  ok = query('start transaction');
  if ( !ok ) raise 'Unable to start transaction!';
  queries = array(
    'lock table output_values in access exclusive mode nowait',
    'truncate output_values_center1',
    'commit'
  );
  if ( i>NOWAIT_TIMES ) {
    // we will wait this time, as we tried NOWAIT_TIMES and failed
    queries[0] = 'lock table output_values in access exclusive mode';
  }
  foreach q in queries {
     ok = query(q);
     if (!ok) break;
  }
  if (!ok) {
    query('rollback');
    usleep(SLEEPTIME_USECS);
  } else {
    break;
  };
};

通过这种方式，您可以安全地避免死锁，因为父表将被独占锁定。用户将在截断运行时阻塞一小部分秒，并在提交后自动恢复。

但要做好准备，这可以在繁忙的服务器上运行几秒钟，因为当表正在使用时，锁将失败并重试。

select和truncate之间的死锁（postgresql）

2 个答案: