Increase minimum page size to 8192

[Adriano dos Santos Fernandes]

Hi!

With schemas, there are new indices and for example in
RDB$PROCEDURE_PARAMETERS there is 4 name fields. That don't fit in
databases with 4096 page size.

Can we increase this minimum in v6?


Adriano

[Dimitry Sibiryakov]

Adriano dos Santos Fernandes wrote 22.06.2024 3:48:
> With schemas, there are new indices and for example in
> RDB$PROCEDURE_PARAMETERS there is 4 name fields.

Why there is schema in parameter names?
Cannot parameters refer procedure by id?

--
WBR, SD.

[Adriano dos Santos Fernandes]

With exception of charset/collation, every object is referenced by name
fields in Firebird system tables.

I don't like it, but it's what we received from the original design.

I don't think this should be changed in a specific place only.

And since removal of very small page sizes isn't probably a big deal, I
think this path should be chosen on this feature.


Adriano

[Dmitry Yemanov]

23.06.2024 03:22, Adriano dos Santos Fernandes пишет:

> With exception of charset/collation, every object is referenced by name
> fields in Firebird system tables.
>
> I don't like it, but it's what we received from the original design.
>
> I don't think this should be changed in a specific place only.

Agreed.

> And since removal of very small page sizes isn't probably a big deal, I
> think this path should be chosen on this feature.

Bigger page sizes are not always good, they have problems too (bigger
latch contention under write-intensive load and higher risks for
database corruption due to partial page writes). But given that 8KB is
the default page size, I don't think things would get any worse overall,
and these issues may (should?) be addressed separately.


Dmitry

[Dimitry Sibiryakov]

Dmitry Yemanov wrote 23.06.2024 8:20:
> 23.06.2024 03:22, Adriano dos Santos Fernandes пишет:
>
>> With exception of charset/collation, every object is referenced by name
>> fields in Firebird system tables.
>>
>> I don't like it, but it's what we received from the original design.
>>
>> I don't think this should be changed in a specific place only.
>
> Agreed.

So we can either a) change it everywhere, or b) wait a little while Vlad
commit indexless unique constraints. IIRC he is working on them, right?

--
WBR, SD.

[Денис Симонов]

So we can either a) change it everywhere, or b) wait a little while Vlad
commit indexless unique constraints. IIRC he is working on them, right?

--
WBR, SD.

There was no talk of any Unique constraint without indexes. There was talk about Foreign Key implementations that may not use an index.