On Thu, Nov 7, 2019 at 10:01 PM Andres Freund <
andres@anarazel.de> wrote:
>
> Hi,
>
> On 2019-11-07 17:03:44 +0530, Amit Kapila wrote:
> > On Thu, Nov 7, 2019 at 4:48 PM Tomas Vondra
> > <
tomas.vondra@2ndquadrant.com> wrote:
> > >
> > > I'm a bit confused - does this happen only with the logical_work_mem
> > > patches, or with clean master too?
> > >
> >
> > This occurs with the clean master. This is a base code problem
> > revealed while doing stress testing of logical_work_mem patches.
>
> As far as I can tell there are no repro steps included? Any chance to
> get those?
>
This problem does not occur consistently. When I was reviewing and testing "logical streaming for large in-progress transactions" link [1] I found the crashes.
This issue does not occur directly, meaning this issue will occur only when some crash occurs in postgres process(not from reorderbuffer but due to some other issue), after the original non-reorderbuffer crash this reorderbuffer crash appears.
To simplify the reorderbuffer crash, I used the following steps:
1) Make replication setup with publisher/subscriber for some table
2) Prepare a sql file with the below:
begin;
4096 insert statements;
select pg_sleep(120)
3) Execute the above script.
4) Attach the postgres process when pg_sleep is in progress.
5) call abort() from attached gdb.
