Heikki Linnakangas <heikki@enterprisedb.com> writes:
>> This patch makes CLUSTER MVCC-safe. Visibility information and update
>> chains are preserved like in VACUUM FULL.
> Here's an update, fixing conflict by Tom's recent commit of Simon's
> patch to skip WAL-inserts when archiving is not enabled.
Applied with revisions. There were some bugs in it: you need to check
both xmin and tid when determining if one tuple chains to another,
and you can't separate MarkBufferDirty from the critical section that
writes xlog. (I got around that by not keeping the working page in
buffers at all, the same way btree index build works; should be a bit
faster as well as more correct.) It had some memory leakage too.
regards, tom lane