Thread: RE: WAL versus Postgres (or: what goes around, comes ar ound)

RE: WAL versus Postgres (or: what goes around, comes ar ound)

From
"Mikheev, Vadim"
Date:
> > I've read this paper ~2 years ago. My plans so far were:
> > 
> > 1. WAL in 7.1
> > 2. New (overwriting) storage manager in 7.2
> > 
> 
> Oh, so Vadim has overwriting storage manager concept for 7.2.
> Vadim, how will you keep old rows around for MVCC?

Just like you told about it - some outstanding files for old
versions. Something like Oracle' rollback segments.
And, for sure, this will be the most complex part of smgr and
that's why I think that we can't use their smgr if we're
going to keep MVCC.

As for WAL, WAL itself (as collection of routines to log changes,
create checkpoints etc) is 90% done. Now it has to be integrated
into system and the most hard part of this work are access methods
specific redo/undo functions. If we're going to use our access
methods then we'll have to write these functions for no matter
what WAL implementation will be used.

Vadim