Thread: Foreign keys
PostGreSql 7.4.3
Am I right in assuming that a foreign key on a table does not explicity create an index on that column on the foreign table? I can see via the \d on the table that the foreign key exists, but accessing the table on that column seems to use a sequential scan.
Am I right in assuming that a foreign key on a table does not explicity create an index on that column on the foreign table? I can see via the \d on the table that the foreign key exists, but accessing the table on that column seems to use a sequential scan.
Regards, Steve Tucknott ReTSol Ltd DDI: 01903 828769 |
Steve Tucknott wrote: > PostGreSql 7.4.3 > > Am I right in assuming that a foreign key on a table does not > explicity create an index on that column on the foreign table? Yes Cheers Noel
On Fri, Aug 27, 2004 at 09:31:51AM +0100, Steve Tucknott wrote: > PostGreSql 7.4.3 > > Am I right in assuming that a foreign key on a table does not explicity > create an index on that column on the foreign table? I can see via the > \d on the table that the foreign key exists, but accessing the table on > that column seems to use a sequential scan. Adding the foreign key won't add the index itself, but I believe that foreign keys can only be declared on columns declared unique. Since declaring columns unique implicitly creates an index on them, if you can legally declare the foreign key then "\d" should show that the foreign key's index already exists. It sounds like sequential scan is being used for a completely unrelated reason. - Jeremy
On Fri, Aug 27, 2004 at 09:31:51 +0100, Steve Tucknott <steve@retsol.co.uk> wrote: > PostGreSql 7.4.3 > > Am I right in assuming that a foreign key on a table does not explicity > create an index on that column on the foreign table? I can see via the > \d on the table that the foreign key exists, but accessing the table on > that column seems to use a sequential scan. Do the data types match?
Jeremy Semeiks <jrs@farviolet.com> writes:
> On Fri, Aug 27, 2004 at 09:31:51AM +0100, Steve Tucknott wrote:
>> Am I right in assuming that a foreign key on a table does not explicity
>> create an index on that column on the foreign table?
> Adding the foreign key won't add the index itself, but I believe that
> foreign keys can only be declared on columns declared unique.
Right, the referenced column must have an index. However, the system
does not require the referencING column to have an index. This can be a
performance loss --- in particular when deleting records from the
referencED table, because the FK machinery then has to seqscan to see
if there are any matching referencING rows. If your master table is
pretty stable, though, you may not care enough to pay the costs of
keeping an index on the slave table.
You can also get burnt if the referenced and referencing columns aren't
of the exact same datatype --- again, not enforced by the system, but a
good way to shoot yourself in the foot performance-wise.
regards, tom lane
Thanks all,
The data types are correct (we use serial/integer fields for the prime and foreign key references).
Where I had gone wrong was in assuming that the slave foreign key declaration created an index on that column on the slave table (it seems to in Informix - where I'm converting from) - I have modified my table create scripts to create the indexes explicitly. Conversion is looking good so far!
On Fri, 2004-08-27 at 20:13, Tom Lane wrote:
The data types are correct (we use serial/integer fields for the prime and foreign key references).
Where I had gone wrong was in assuming that the slave foreign key declaration created an index on that column on the slave table (it seems to in Informix - where I'm converting from) - I have modified my table create scripts to create the indexes explicitly. Conversion is looking good so far!
On Fri, 2004-08-27 at 20:13, Tom Lane wrote:
Jeremy Semeiks <jrs@farviolet.com> writes: > On Fri, Aug 27, 2004 at 09:31:51AM +0100, Steve Tucknott wrote: >> Am I right in assuming that a foreign key on a table does not explicity >> create an index on that column on the foreign table? > Adding the foreign key won't add the index itself, but I believe that > foreign keys can only be declared on columns declared unique. Right, the referenced column must have an index. However, the system does not require the referencING column to have an index. This can be a performance loss --- in particular when deleting records from the referencED table, because the FK machinery then has to seqscan to see if there are any matching referencING rows. If your master table is pretty stable, though, you may not care enough to pay the costs of keeping an index on the slave table. You can also get burnt if the referenced and referencing columns aren't of the exact same datatype --- again, not enforced by the system, but a good way to shoot yourself in the foot performance-wise. regards, tom lane ---------------------------(end of broadcast)--------------------------- TIP 9: the planner will ignore your desire to choose an index scan if your joining column's datatypes do not match
Regards, Steve Tucknott ReTSol Ltd DDI: 01903 828769 |