Re: BUG #13541: There is a visibility issue when run some DDL and Query. The time window is very shot - Mailing list pgsql-bugs

From Haribabu Kommi
Subject Re: BUG #13541: There is a visibility issue when run some DDL and Query. The time window is very shot
Date
Msg-id CAJrrPGd--dTUoRpwiSmNVpa56sbCMMB-xB4=Qc4um1NNnt4uhQ@mail.gmail.com
Whole thread Raw
In response to BUG #13541: There is a visibility issue when run some DDL and Query. The time window is very shot  (beijing_pg@163.com)
Responses Re: BUG #13541: There is a visibility issue when run some DDL and Query. The time window is very shot  (zhangjinyu <beijing_pg@163.com>)
List pgsql-bugs
On Fri, Aug 7, 2015 at 10:16 PM,  <beijing_pg@163.com> wrote:
> The following bug has been logged on the website:
>
> Bug reference:      13541
> Logged by:          zhangjinyu
> Email address:      beijing_pg@163.com
> PostgreSQL version: 9.2.13
> Operating system:   suse
> Description:
>
> There is a visibility issue when run some DDL and Query. The time window is
> very shot. Postgres is to hold relation lock after getsnapshot. Before a
> query holds relation lock, if another session run a DDL(alter table alter
> column set datatype), Once this DDL committed, select query will return zero
> row, because the DDL(alter table alter column set datatype) will rewrite all
> tuples with new transaction id.
>
> We can use the following test case to repro the issue.  we add sleep(40)
> after GetTransactionSnapshot (in function exec_simple_query) to enlarge the
> time window.
> Step1: session1: select * from t; (return 3 row)
>                  set enable_sleep30_after_getsnapshot=on;
> select * from t;
> Step2: session2: begin transaction;
> alter table t alter c1 set type char(10);
> commit;
>
> we can see session1 "select * from t" return 0 row.
> ======================================
> session1:
> select * from t;  (it returns 3 rows)
> set enable_sleep30_after_getsnapshot=on;
> select * from t;  (it returns 0 row)
>
> session2:
> start transaction;
> alter table t alter c1 set type char(10);
> end;

I am not able to reproduce this issue.
I tried putting a breakpoint after GetTransactionSnapshot function
in exec_simple_query function and executed the alter table in other session.

Even if the other session data is committed, still the old data is
present in the
database. The old transaction still can see this data. Am I missing something?

Regards,
Hari Babu
Fujitsu Australia

pgsql-bugs by date:

Previous
From: Sushant Sinha
Date:
Subject: Re: BUG #13515: Much higher disk writes after postgres upgrade 9.4.3->9.4.
Next
From: Michael Meskes
Date:
Subject: Re: BUG #13557: ESQL select issue