On 10/10/22 8:12 AM, Kostas Papadopoulos wrote:
On 10/10/2022 17:53, Tom Lane wrote:
Kostas Papadopoulos <kostas@xxxxxxxxxxxxxxxxx> writes:
I cannot see how it can be configuration since the two databases are
in the same
Postgres instance.
There is such a thing as ALTER DATABASE ... SET to install different
settings at the per-database level.
I understand, but I created the databases to be the same. Our original
problem was that developers' workstations (Debian and Windows) were
running a specific query different from a test db (Ubuntu). After
eliminating everything we thought of (data, versions, configurations,
OS, etc) we ended up with the scenario I described here.
So there is more to the story.
Information needed:
1) The query and its EXPLAIN ANALYZE for both slow/fast cases.
2) Postgres version.
3) What database are the developers workstation pointing at?
4) What is the test db and is it the same as 3)?
5) What clients are you using to run the query?
In general, the answer to your question is that the databases are
*not* identical. You just haven't figured out how yet. I'm wondering
if it has something to do with the dump/reload having compacted out
bloat in the tables or indexes, causing cost estimates to change.
I will look into that and a couple of other ideas I got from this list.
regards, tom lane
Thanks
kostas
--
Adrian Klaver
adrian.klaver@xxxxxxxxxxx
