From: | Heikki Linnakangas <hlinnakangas(at)vmware(dot)com> |
---|---|
To: | Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> |
Cc: | PostgreSQL-development <pgsql-hackers(at)postgresql(dot)org> |
Subject: | Re: Gin page deletion bug |
Date: | 2013-11-08 20:33:24 |
Message-ID: | 527D4A94.5050308@vmware.com |
Views: | Raw Message | Whole Thread | Download mbox | Resend email |
Thread: | |
Lists: | pgsql-hackers |
On 08.11.2013 19:14, Tom Lane wrote:
> Heikki Linnakangas <hlinnakangas(at)vmware(dot)com> writes:
>> I came up with the attached fix. In a nutshell, when walking along a
>> right-link, the new page is locked before releasing the lock on the old
>> one. Also, never delete the leftmost branch of a posting tree. I believe
>> these changes are sufficient to fix the problem, because of the way the
>> posting tree is searched:
>
> This seems reasonable, but I wonder whether the concurrency discussion
> shouldn't be in gist/README, rather than buried in a comment inside
> ginDeletePage (not exactly the first place one would think to look IMO).
Yeah, README is probably better. There's zero explanation of concurrency
and locking issues there at the moment, so I added a new section there
explaining why the page deletion is (now) safe. There's a lot more to
say about locking in GIN, but I'll leave that for another day.
- Heikki
From | Date | Subject | |
---|---|---|---|
Next Message | Alvaro Herrera | 2013-11-08 20:36:02 | Re: Minmax indexes |
Previous Message | Tom Lane | 2013-11-08 20:21:29 | Re: regclass error reports improperly downcased |