Thread: GetTransactionSnapshot() in enum.c
Hi, ISTM that we shouldn't use GetTransactionSnapshot() in enum.c but GetLatestSnapshot() in <= 9.3 and NULL/GetCatalogSnapshot() > 9.3. typecache.c's usage was converted to GetLatestSnapshot() but enum.c's was not. I don't seem to have full mental capacity right now, but I think the current usage could cause problems with a range type index over a enum column. Index predicates might also be problematic. Greetings, Andres Freund -- Andres Freund http://www.2ndQuadrant.com/PostgreSQL Development, 24x7 Support, Training & Services
Andres Freund <andres@2ndquadrant.com> writes:
> ISTM that we shouldn't use GetTransactionSnapshot() in enum.c but
> GetLatestSnapshot() in <= 9.3 and NULL/GetCatalogSnapshot() > 9.3.
> typecache.c's usage was converted to GetLatestSnapshot() but enum.c's
> was not.
That was intentional, see the comments for commit
9ad45c18b6c8d03ce18a26223eb0d15e900c7a2c.
Possibly we should rethink this in HEAD given that we don't do SnapshotNow
scans anymore, but I'm disinclined to do so in back branches.
BTW, I notice that the MVCC-catalog-scans patch summarily asserts that
RenumberEnumType no longer poses any concurrency hazards. I doubt that's
true: isn't it still possible that pg_enum rows acquired through the
syscaches will have inconsistent enumsortorder values, if they were
read at different times? If you want to examine enumsortorder, you really
need to be comparing rows you know were read with the *same* snapshot.
regards, tom lane
On Mon, Aug 19, 2013 at 1:41 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Andres Freund <andres@2ndquadrant.com> writes: >> ISTM that we shouldn't use GetTransactionSnapshot() in enum.c but >> GetLatestSnapshot() in <= 9.3 and NULL/GetCatalogSnapshot() > 9.3. > >> typecache.c's usage was converted to GetLatestSnapshot() but enum.c's >> was not. > > That was intentional, see the comments for commit > 9ad45c18b6c8d03ce18a26223eb0d15e900c7a2c. > > Possibly we should rethink this in HEAD given that we don't do SnapshotNow > scans anymore, but I'm disinclined to do so in back branches. > > BTW, I notice that the MVCC-catalog-scans patch summarily asserts that > RenumberEnumType no longer poses any concurrency hazards. I doubt that's > true: isn't it still possible that pg_enum rows acquired through the > syscaches will have inconsistent enumsortorder values, if they were > read at different times? If you want to examine enumsortorder, you really > need to be comparing rows you know were read with the *same* snapshot. Good point, I missed that. Here's a proposed comment patch. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company
Attachment
Robert Haas <robertmhaas@gmail.com> writes:
> On Mon, Aug 19, 2013 at 1:41 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
>> BTW, I notice that the MVCC-catalog-scans patch summarily asserts that
>> RenumberEnumType no longer poses any concurrency hazards. I doubt that's
>> true: isn't it still possible that pg_enum rows acquired through the
>> syscaches will have inconsistent enumsortorder values, if they were
>> read at different times? If you want to examine enumsortorder, you really
>> need to be comparing rows you know were read with the *same* snapshot.
> Good point, I missed that. Here's a proposed comment patch.
Looks sane to me.
regards, tom lane
On Mon, Aug 26, 2013 at 4:31 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: > Robert Haas <robertmhaas@gmail.com> writes: >> On Mon, Aug 19, 2013 at 1:41 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote: >>> BTW, I notice that the MVCC-catalog-scans patch summarily asserts that >>> RenumberEnumType no longer poses any concurrency hazards. I doubt that's >>> true: isn't it still possible that pg_enum rows acquired through the >>> syscaches will have inconsistent enumsortorder values, if they were >>> read at different times? If you want to examine enumsortorder, you really >>> need to be comparing rows you know were read with the *same* snapshot. > >> Good point, I missed that. Here's a proposed comment patch. > > Looks sane to me. Thanks for the review. Committed. -- Robert Haas EnterpriseDB: http://www.enterprisedb.com The Enterprise PostgreSQL Company