Обсуждение: Re: [HACKERS] select order by for update

From: Vadim Mikheev <vadim@krs.ru>

|> There is another problem in subj - sometimes application gets
|> ERROR: EvalPlanQual: t_xmin is uncommitted ?!
|> I'll try to find why. Mark (Wright), could you avoid
|> order by in PL function? If you really need in ordered
|> updates then try to create index on id_number and add
|> id_number >= 0 to WHERE in select for update.
|
|Ops, this will not work in all cases. Try to rewrite select:
|
|SELECT * FROM test_attendees
|WHERE print_status = 'R'
|AND id_number = (select min(id_number) from test_attendees)
|FOR UPDATE OF test_attendees
|
|and run it in loop.

That would only work the first time, since after updating print_status to
'C', the record where id_number = (select min(id_number) from
test_attendees) would no longer have print_status = 'R', so no records would
match the query.

The solution would appear to be to replace the clause '(select
min(id_number) from test_attendees)' with '(select min(id_number) from
test_attendees where print_status = 'R')'.  However, that would not work,
since the subselect doesn't block (see the pgsql mailing list for an
explanation from Jan Wieck -
http://www.postgresql.org/mhonarc/pgsql-sql/1999-06/msg00049.html - my
current solution is from his suggestion).

I need the ordered select, since I'm trying to create a FIFO.  I have one
set of clients who are entering records into the table, and another set of
clients who are taking those records and sending them to a printer.  I need
the printers to output records in more or less the same order that they were
entered.

---
Mark Wright
mwright@pro-ns.net
mark_wright@datacard.com

Mark Wright wrote:
> 
> |SELECT * FROM test_attendees
> |WHERE print_status = 'R'
> |AND id_number = (select min(id_number) from test_attendees  )
   ^^^
 
Ops - add "AND print_status = 'R'" here!

> |FOR UPDATE OF test_attendees
> |
> |and run it in loop.
> 
> That would only work the first time, since after updating print_status to
> 'C', the record where id_number = (select min(id_number) from
> test_attendees) would no longer have print_status = 'R', so no records would
> match the query.

And this is why I said "run select in loop": if no one row will
be returned (i.e. row returned by subselect is already updated
by concurrent xaction) then re-run select!

Vadim