Обсуждение: READ UNCOMMITTED in postgres

Greetings,

* Matthew Phillips (mphillips34@gmail.com) wrote:
> With the current READ UNCOMMITTED discussion happening on pgsql-hackers
> [1], It did raise a question/use-case I recently encountered and could not
> find a satisfactory solution for. If someone is attempting to poll for new
> records on a high insert volume table that has a monotonically increasing
> id, what is the best way to do it? As is, with a nave implementation, rows
> are not guaranteed to appear in monotonic order; so if you were to keep a
> $MAX_ID, and SELECT WHERE p_id > $MAX_ID, you would hit gaps. Is there a
> clean way to do this? I've seen READ UNCOMMITTED used for this with DB2.

There's the LISTEN/NOTIFY system, which at a high level is a better
approach than using a polling system.

Thanks,

Stephen

Matthew Phillips schrieb am 19.12.2019 um 00:12:
> Hi, With the current READ UNCOMMITTED discussion happening on
> pgsql-hackers [1], It did raise a question/use-case I recently
> encountered and could not find a satisfactory solution for. If
> someone is attempting to poll for new records on a high insert volume
> table that has a monotonically increasing id, what is the best way to
> do it? As is, with a nave implementation, rows are not guaranteed to
> appear in monotonic order; so if you were to keep a $MAX_ID, and
> SELECT WHERE p_id > $MAX_ID, you would hit gaps. Is there a clean way
> to do this? I've seen READ UNCOMMITTED used for this with DB2.

In my understanding READ UNCOMMITTED in other databases is typically used to avoid read-locks which Postgres doesn't
have.
 
So I wonder what benefits READ UNCOMMITTED would have to begin with.

But, if you want to poll for new rows, then why don't you use a timestamp column?

  select *
  from the_table
  where created_at >= <last check time>