Re: Relation extension scalability

On Tue, Mar 29, 2016 at 1:29 PM, Dilip Kumar <dilipbalaut(at)gmail(dot)com> wrote:
> Attaching new version v18
>
> - Some cleanup work on v17.
> - Improved UpdateFreeSpaceMap function.
> - Performance and space utilization are same as V17

Looks better. Here's a v19 that I hacked on a bit.

Unfortunately, one compiler I tried this with had a pretty legitimate complaint:

hio.c: In function ‘RelationGetBufferForTuple’:
hio.c:231:20: error: ‘freespace’ may be used uninitialized in this
function [-Werror=uninitialized]
hio.c:185:7: note: ‘freespace’ was declared here
hio.c:231:20: error: ‘blockNum’ may be used uninitialized in this
function [-Werror=uninitialized]
hio.c:181:14: note: ‘blockNum’ was declared here

There's nothing whatsoever to prevent RelationExtensionLockWaiterCount
from returning 0.

It's also rather ugly that the call to UpdateFreeSpaceMap() assumes
that the last value returned by PageGetHeapFreeSpace() is as good as
any other, but maybe we can just install a comment explaining that
point; there's not an obviously better approach that I can see.

--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company