Re: Calling PLpgSQL function with composite type fails with an error:"ERROR: could not open relation with OID ..."

Поиск
Список
Период
Сортировка
От Ashutosh Sharma
Тема Re: Calling PLpgSQL function with composite type fails with an error:"ERROR: could not open relation with OID ..."
Дата
Msg-id CAE9k0PnjG_O+Lh=Aq41S=MD820oVKAmb3koLruMzqBwgmCrF8w@mail.gmail.com
обсуждение исходный текст
Ответ на Calling PLpgSQL function with composite type fails with an error:"ERROR: could not open relation with OID ..."  (Ashutosh Sharma <ashu.coek88@gmail.com>)
Ответы Re: Calling PLpgSQL function with composite type fails with an error: "ERROR: could not open relation with OID ..."  (Tom Lane <tgl@sss.pgh.pa.us>)
Список pgsql-hackers
The issue here is that PLpgSQL_rec structure being updated by
revalidate_rectypeid() is actually a local/duplicate copy of the
PLpgSQL_rec structure available in plpgsql_HashTable (refer to
copy_plpgsql_datums() where you would notice that if datum type is
PLPGSQL_DTYPE_REC we actually mempcy() the PLpgSQL_rec structure
available in func >datums[] array). This basically means that the
rectypeid field updated post typcache entry validation in
revalidation_rectypeid() is actually a field in duplicate copy of
PLpgSQL_rec structure, not the original copy of it available in
func->datums[]. Hence, when the same function is executed for the
second time, the rectypeid field of PLpgSQL_rec structure being
reloaded from the func->datums[] actually contains the stale value
however the typcache entry in it is up-to-date which means
revalidation_rectypeid() returns immediately leaving a stale value in
rectypeid. This causes the function make_expanded_record_from_typeid()
to use the outdated value in rec->rectypeid resulting into the given
error.

To fix this, I think instead of using rec->rectypeid field we should
try using rec->datatype->typoid when calling
make_expanded_record_from_typeid(). Here is the change that I'm
suggesting:

--- a/src/pl/plpgsql/src/pl_exec.c
+++ b/src/pl/plpgsql/src/pl_exec.c
@@ -6942,7 +6942,7 @@ make_expanded_record_for_rec(PLpgSQL_execstate *estate,
            newerh = make_expanded_record_from_exprecord(srcerh,
                                                         mcontext);
        else
-           newerh = make_expanded_record_from_typeid(rec->rectypeid, -1,
+          newerh = make_expanded_record_from_typeid(rec->datatype->typoid, -1,

Thoughts ?

--
With Regards,
Ashutosh Sharma
EnterpriseDB:http://www.enterprisedb.com

On Thu, Dec 26, 2019 at 3:57 PM Ashutosh Sharma <ashu.coek88@gmail.com> wrote:
>
> Hi All,
>
> When the following test-case is executed on master, it fails with an
> error: "ERROR:  could not open relation with OID ..."
>
> -- create a test table:
> create table tab1(a int, b text);
>
> -- create a test function:
> create or replace function f1() returns void as
> $$
> declare
>    var1 tab1;
> begin
>    select * into var1 from tab1;
> end
> $$ language plpgsql;
>
> -- call the test function:
> select f1();
>
> -- drop the test table and re-create it:
> drop table tab1;
> create table tab1(a int, b text);
>
> -- call the test function:
> select f1();
>
> -- call the test function once again:
> select f1(); -- this fails with an error "ERROR:  could not open
> relation with OID .."
>
> I'm trying to investigate this issue and will try to share my findings soon...
>
> --
> With Regards,
> Ashutosh Sharma
> EnterpriseDB:http://www.enterprisedb.com



В списке pgsql-hackers по дате отправления:

Предыдущее
От: Sergei Kornilov
Дата:
Сообщение: Re: Expose lock group leader pid in pg_stat_activity
Следующее
От: Julien Rouhaud
Дата:
Сообщение: Re: Expose lock group leader pid in pg_stat_activity