Thread

1. Re: Bugs in CREATE/DROP INDEX CONCURRENTLY

   Kevin Grittner <kgrittn@mail.com> — 2012-10-17T22:51:06Z

   Kevin Grittner wrote:
   > Simon Riggs wrote:
   >> On 6 October 2012 00:56, Tom Lane <tgl@sss.pgh.pa.us> wrote:
   >  
   >>> 2. DROP INDEX CONCURRENTLY doesn't bother to do
   >>> TransferPredicateLocksToHeapRelation until long after it's
   >>> invalidated the index. Surely that's no good? Is it even possible
   >>> to do that correctly, when we don't have a lock that will prevent
   >>> new predicate locks from being taken out meanwhile?
   >> 
   >> No idea there. Input appreciated.
    
   > If the creation of a new tuple by insert or update would not
   > perform the related index tuple insertion, and the lock has not yet
   > been transfered to the heap relation, yes we have a problem.  Will
   > take a look at the code.
   > 
   > Creation of new predicate locks while in this state has no bearing
   > on the issue as long as locks are transferred to the heap relation
   > after the last scan using the index has completed.
    
   To put that another way, it should be done at a time when it is sure
   that no query sees indisvalid = true and no query has yet seen
   indisready = false.  Patch attached.  Will apply if nobody sees a
   problem with it.
   
   -Kevin
   

2. Re: Bugs in CREATE/DROP INDEX CONCURRENTLY

   Tom Lane <tgl@sss.pgh.pa.us> — 2012-10-17T23:31:31Z

   "Kevin Grittner" <kgrittn@mail.com> writes:
   > To put that another way, it should be done at a time when it is sure
   > that no query sees indisvalid = true and no query has yet seen
   > indisready = false. Patch attached. Will apply if nobody sees a
   > problem with it.
   
   The above statement of the requirement doesn't seem to match what you
   put in the comment:
   
   > + 	 * All predicate locks on the index are about to be made invalid. Promote
   > + 	 * them to relation locks on the heap. For correctness this must be done
   > + 	 * after the index was last seen with indisready = true and before it is
   > + 	 * seen with indisvalid = false.
   
   and the comment is rather vaguely worded too (last seen by what?).
   Please wordsmith that a bit more.
   
   			regards, tom lane