_bt_pgaddtup, called by _bt_split, always sets the LP_USED flag
and after that the following transactions access the heap data (which resides on
disk) to check whether the tuple is alive. But this is unnessesary when the
original tuple is not alive, because in this case the new item is necessarily
not alive too. If we passed down the status flag (LP_DELETE/LP_USED), we would
know this and could save one disk access.
I would appreciate your comments on this.
Toru SHIMOGAKI <shimogakit(at)intellilink(dot)co(dot)jp>
pgsql-hackers by date
|Next:||From: Poul Jensen||Date: 2006-01-30 04:34:19|
|Subject: Re: Failed install - libgen.so doesn't exist|
|Previous:||From: Tom Lane||Date: 2006-01-30 04:00:39|
|Subject: Re: Failed install - libgen.so doesn't exist |