Alvaro Herrera wrote:
However, given that Heikki just confirmed that CLUSTER does not freeze
tuples, it's not really possible to do this, so I'll drop the CLUSTER
patch for now.

This means that people using CLUSTER to compact tables won't have the
benefit of advancing relfrozenxid, so they will have to run VACUUM on
those tables at some point anyway, even though there will be no dead
tuples :-(

Now that I think this a bit more, I think CLUSTER should freeze the tuples. I was worried about losing valuable debug information by doing that, but thinking a bit more that's not a big concern: we wouldn't freeze tuples newer than recent xmin. The update chain logic is the most risky part of the code, and we wouldn't lose the xmin and xmax of tuples that are part of update chains.

Patch attached. You'll need the changes to cluster.c to set the relfrozenid as well.

--
  Heikki Linnakangas
  EnterpriseDB   http://www.enterprisedb.com
Index: src/backend/access/heap/rewriteheap.c
===================================================================
RCS file: /home/hlinnaka/pgcvsrepository/pgsql/src/backend/access/heap/rewriteheap.c,v
retrieving revision 1.3
diff -c -r1.3 rewriteheap.c
*** src/backend/access/heap/rewriteheap.c	17 Apr 2007 21:29:31 -0000	1.3
--- src/backend/access/heap/rewriteheap.c	16 May 2007 13:42:23 -0000
***************
*** 538,543 ****
--- 538,545 ----
  	OffsetNumber	newoff;
  	HeapTuple		heaptup;
  
+ 	heap_freeze_tuple(tup->t_data, state->rs_oldest_xmin, NULL);
+ 
  	/*
  	 * If the new tuple is too big for storage or contains already toasted
  	 * out-of-line attributes from some other relation, invoke the toaster.
---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
       choose an index scan if your joining column's datatypes do not
       match

Reply via email to