On Thu, Jun 16, 2011 at 6:54 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I believe that this is fundamentally unavoidable so long as we use
> SnapshotNow to read catalogs --- which is something we've talked about
> changing, but it will require a pretty major R&D effort to make it
> happen. In the meantime, we have to go back to using
> AccessExclusiveLock for table alterations. It doesn't help to have
> a lower lock level if that means that concurrent transactions will
> unpredictably fail instead of waiting.
Ouch.
I wonder if we could avoid this anomaly by taking a throwaway MVCC
snapshot at the beginning of each system catalog scan and using it
just for the duration of that scan. If nothing that has touched the
catalog commits while the scan is open, then this is logically
equivalent to SnapshotNow. If something does commit in mid-scan, then
we might not get the latest version of the row, but we should end up
with exactly one. If it's not the latest one, we'll do the rebuild
again upon seeing the next sinval message; in the meantime, the
version we're using mustn't be too intolerably bad or it was an error
not to use AccessExclusiveLock in the first place.
IIUC, the problem with this approach is not correctness but
performance. Taking snapshots is (currently) expensive.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company