Tom Lane wrote:
Rajesh Kumar Mallah <mallah@trade-india.com> writes:
Yep the problem of original posting could be replicated on
disabling hash aggregates. On disabling i could get the repeated rows.
Okay. What I suspect is happening is that there are entries in the
column that are equal according to the datatype's comparison function,
but are not bitwise equal and therefore yield different hash codes.
This makes it a crapshoot whether they are seen to be equal or not
when hash aggregation is used. We identified a similar bug in the
inet/cidr datatypes just a few weeks ago.
What exactly is the datatype of the "name" column?
name | character varying(120) | not null If it's a text
type, what database encoding and locale settings (LC_COLLATE/LC_CTYPE)
are you using?
List of databases
+-----------------+----------+-----------+
| Name | Owner | Encoding |
+-----------------+----------+-----------+
| bric | postgres | UNICODE |
+--------------------------------+-----------------+
| name | setting |
+--------------------------------+-----------------+
| lc_collate | en_US.UTF-8 |
| lc_ctype | en_US.UTF-8 |
| lc_messages | en_US.iso885915 |
| lc_monetary | en_US.iso885915 |
| lc_numeric | en_US.iso885915 |
| lc_time | en_US.iso885915 |
Can you investigate exactly what's stored within each
of these groups of matching names?
Can you tell me how to do it please?
regards, tom lane
Regds
Mallah.
