Re: error "can only drop stats once" brings down database

At Mon, 17 Jun 2024 08:28:10 +0900, Michael Paquier <michael@paquier.xyz> wrote in 
> On Fri, Jun 14, 2024 at 09:37:35AM +0900, Michael Paquier wrote:
> > I am wondering if we should tackle this problem by adding a notion of
> > "age" in the stats entries for what's stored in the central dshash and
> > in the hold copies in the backends, so as we dump the "age" of an
> > existing stats entry if we find out that it gets reused in a create
> > path.  That requires some design and would not be backpatchable, but
> > it does not seem that invasive at quick glance (I had my eyes on this
> > code the full week).  Other thoughts are welcome.
> 
> This still sounds like a good idea after sleeping on it over the
> weekend.  Please note that I am moving on with the problem on this
> thread where this has been reported, and that is the point of origin:
> https://www.postgresql.org/message-id/ZmvVudpT8z6hbY9t@paquier.xyz
> 
> Let's continue the discussion there.

Is it something like this? I forgot the reason why this was excluded
from the vinal version, though.

https://www.postgresql.org/message-id/20201001.090722.322196928507744460.horikyota.ntt%40gmail.com

> Dshash scans are used for dropping and resetting stats entries. Entry
> dropping is performed in the following steps.
> 
> (delete_current_stats_entry())
> - Drop the dshash entry (needs exlock of dshash partition).
> 
> - If refcount of the stats entry body is already zero, free the memory
>    immediately .
> 
> - If not, set the "dropped" flag of the body. No lock is required
>   because the "dropped" flag won't be even referred to by other
>   backends until the next step is done.
> 
> - Increment deletion count of the shared hash. (this is used as the
>   "age" of local pointer cache hash (pgstat_cache).
> 
> (get_stat_entry())
> 
> - If dshash deletion count is different from the local cache age, scan
>   over the local cache hash to find "dropped" entries.
> 
> - Decrements refcount of the dropped entry and free the shared entry
>   if it is no longer referenced. Apparently no lock is required.

regards.

-- 
Kyotaro Horiguchi
NTT Open Source Software Center