Обсуждение: v8.1 pgbench ERRORs

Поиск
Список
Период
Сортировка

v8.1 pgbench ERRORs

От
"Eric B. Ridge"
Дата:
Okay, don't flame me for (still) using Postgres v8.1.10, but we were
running pgbench on new hardware today, and in the middle of like 50
various runs of pgbench, we saw this:

pgbench -c 20 -t 500 pgbench
starting vacuum...end.
Client 4 aborted in state 8: ERROR:  integer out of range
Client 3 aborted in state 8: ERROR:  integer out of range
Client 17 aborted in state 8: ERROR:  integer out of range
Client 1 aborted in state 8: ERROR:  integer out of range
Client 13 aborted in state 8: ERROR:  integer out of range
Client 7 aborted in state 8: ERROR:  integer out of range
Client 2 aborted in state 8: ERROR:  integer out of range
Client 19 aborted in state 8: ERROR:  integer out of range
Client 9 aborted in state 8: ERROR:  integer out of range
Client 16 aborted in state 8: ERROR:  integer out of range
Client 12 aborted in state 8: ERROR:  integer out of range
Client 15 aborted in state 8: ERROR:  integer out of range
Client 14 aborted in state 8: ERROR:  integer out of range
Client 18 aborted in state 8: ERROR:  integer out of range
Client 5 aborted in state 8: ERROR:  integer out of range
Client 10 aborted in state 8: ERROR:  integer out of range
Client 6 aborted in state 8: ERROR:  integer out of range
Client 8 aborted in state 8: ERROR:  integer out of range
Client 11 aborted in state 8: ERROR:  integer out of range
Client 0 aborted in state 8: ERROR:  integer out of range
transaction type: TPC-B (sort of)
scaling factor: 20
number of clients: 20
number of transactions per client: 500
number of transactions actually processed: 305/10000
tps = 734.580434 (including connections establishing)
tps = 819.883657 (excluding connections establishing)

After reinitializing the pgbench database, everything is running fine.

Anyone seen this before?  Could it really be related to an old version
of PG, whose source isn't even downloadable anymore?  Or would this
point to hardware problems?

eric

Re: v8.1 pgbench ERRORs

От
Tom Lane
Дата:
"Eric B. Ridge" <ebr@tcdi.com> writes:
> Anyone seen this before?

Yeah, you'll eventually overflow the balance fields because the deltas
are always positive.  More recent versions of pgbench use a
positive-and-negative range for delta.

            regards, tom lane

Re: v8.1 pgbench ERRORs

От
"Eric B. Ridge"
Дата:
On Sep 24, 2009, at 5:15 PM, Tom Lane wrote:

> Yeah, you'll eventually overflow the balance fields because the deltas
> are always positive.  More recent versions of pgbench use a
> positive-and-negative range for delta.

Awesome, thanks!  Was really worried that our new hardware was flaking
out.

eric