Simon Liesenfeld wrote:
> Before Diving into the new partitioning methods, of which I assume, that they imply constraints on the used data
model,
> I observed, that a table is a file, e.g. data/base/1234/5678 . As soon the files size exceeds
> 1 Gig, more files will be created, 5678.1 for the next Gig, 5678.2 for another , and so on.
> I had 10 of them for my CTE test table.
>
> I brought down the pg-server, moved the ones with odd extension number to another disk and created
> symbolic links in data/base/1234/ pointing to the moved files.
> I restarted the pg-server, I was surprised, that everything still worked well, and I think,
> I noticed some speedup of my query. I assume, that is, because pg gathers the data of the files
> in parallel, and two disk can shuffle more data in the same time than one disk.
>
> Is this a recommendable practice? Is it supported by postgres to do this by CREATE TABLE, Alter table?
> Can somebody confirm the performance gain, does it even improve with more disks?
No, it is not recommended, it will break your database.
Don't manually mess with the table files.
If you want to place some partitions on a different disk, create a tablespace
on that disk and ALTER the tables to move them there.
Depending on your system and query, placing different partitions on
different disks can improve your performance, particularly if I/O is
the bottleneck.
Yours,
Laurenz Albe
--
Cybertec | https://www.cybertec-postgresql.com
