Re: fixing old_snapshot_threshold's time->xid mapping

Robert Haas <robertmhaas@gmail.com>

From: Robert Haas <robertmhaas@gmail.com>
To: Dilip Kumar <dilipbalaut@gmail.com>
Cc: Kevin Grittner <kgrittn@gmail.com>, Andres Freund <andres@anarazel.de>, "pgsql-hackers@postgresql.org" <pgsql-hackers@postgresql.org>
Date: 2020-04-20T18:01:07Z
Lists: pgsql-hackers
On Mon, Apr 20, 2020 at 12:10 AM Dilip Kumar <dilipbalaut@gmail.com> wrote:
> I have started reviewing these patches.  I think, the fixes looks right to me.
>
> + LWLockAcquire(OldSnapshotTimeMapLock, LW_SHARED);
> + mapping->head_offset = oldSnapshotControl->head_offset;
> + mapping->head_timestamp = oldSnapshotControl->head_timestamp;
> + mapping->count_used = oldSnapshotControl->count_used;
> + for (int i = 0; i < OLD_SNAPSHOT_TIME_MAP_ENTRIES; ++i)
> + mapping->xid_by_minute[i] = oldSnapshotControl->xid_by_minute[i];
> + LWLockRelease(OldSnapshotTimeMapLock);
>
> I think memcpy would be a better choice instead of looping it for all
> the entries, since we are doing this under a lock?

When I did it that way, it complained about "const" and I couldn't
immediately figure out how to fix it.

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



Commits

  1. Remove the "snapshot too old" feature.

  2. Fix two bugs in MaintainOldSnapshotTimeMapping.

  3. Add new 'old_snapshot' contrib module.

  4. Expose oldSnapshotControl definition via new header.