From: David E. W. <da...@ju...> - 2014-02-04 01:15:35
|
On Feb 3, 2014, at 5:06 PM, Koichi Suzuki <koi...@gm...> wrote: > When you turn off enforce_two_phase_commit to off, you should be > careful that if a transactions is involved with one node (coordinator > and/or datanode), commit is not associated with prepare transaction > and you may need manual database recovery if one of the commits fails. > > We're using this in the regression test mainly to maintain temporary > object usage. We can distinguish such implicit 2PC from explicit > ones so we may be able to enable TEMP objects in such case in 1.3 or > later. Yeah, I just won’t ever be able to run pgTAP until proper temporary table and SAVEPOINT support is in. I can live with this, as I can always run my tests against Postgres, even if we deploy to XC. Should get to loading more data into my test cluster this week or next! Thanks, David |