On Mon, Mar 17, 2014 at 10:54 AM, Heikki Linnakangas
<hlinnakangas@vmware.com> wrote:
> Heap and B-tree WAL records also rely on PageAddItem etc. to reconstruct the
> page, instead of making a physical copy of the modified parts. And
> _bt_restore_page even inserts the items physically in different order than
> the normal codepath does. So for good or bad, there is some precedence for
> this.
Yikes.
> The imminent danger I see is if we change the logic on how the items are
> divided into posting lists, and end up in a situation where a master server
> adds an item to a page, and it just fits, but with the compression logic the
> standby version has, it cannot make it fit. As an escape hatch for that, we
> could have the WAL replay code try the compression again, with a larger max.
> posting list size, if it doesn't fit at first. And/or always leave something
> like 10 bytes of free space on every data page to make up for small
> differences in the logic.
That scares the crap out of me. I don't see any intrinsic problem
with relying on the existence page contents to figure out how to roll
forward, as PageAddItem does; after all, we do FPIs precisely so that
the page is in a known good state when we start. However, I really
think we ought to try hard to make this deterministic in terms of what
the resulting state of the page is; anything else seems like it's
playing with fire, and I bet we'll get burned sooner rather than
later.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company