Re: WIP checksums patch

> On 10/1/12 12:22 PM, Josh Berkus wrote:
>>> >Perhaps we don't allow this to be turned per page, but rather per
>>> >cluster, and per-cluster would require the entire cluster to be
>>> >rewritten.
>> We dicussed this last year, and options which require a total rewrite of
>> the database in order to turn on the option were rejected as impractical
>> for users.
>
> For whatever it's worth... we (and presumably others) still use londiste
> (or Slony) as our upgrade path, so we could tolerate a cluster-wide
> setting. We'd just set it when building new clusters via londiste and
> forget about it.
>
> So I'd rather see this get in at a cluster level than not make it at all
> while we wait for something better.

Some still do dump/restore between major releases, so there it is also
applicable. (and preferrable).

I do know a lot of things had been discussed last year, an API I could
easily imagine would work:

ALTER TABLE <tablename> SET ENABLE_CHECKSUMMING=YES
(would make the server do the checksums on all writes on table+indices
going forward, here, verification of the checksums would still be enabled,
but only on "already checksummed pages" ).
ALTER TABLE <tablename> set VERIFY_CHECKSUM=YES
(would cause the server to scan table and index, and rewrite the parts
that hadn't an updated with a checksum already.

Perhaps the names are not well-chosen, but it is based on the assumptions
that check-summing overhead is measurable and it is thereby desireable to
be able to dis/enable it on a per-table level.

Then I could let my system go 6-12 months between the above two commands
and the amount of rewrite would hopefully not require a rewrite of the
entire 2.5TB of PGDATA.

Jesper

--
Jesper Krogh