On Thu, Aug 22, 2019 at 12:47 AM Alvaro Herrera <alvhe...@2ndquadrant.com>
wrote:
>
> On 2019-Aug-21, Melanie Plageman wrote:
>
> > In Greenplum, we mainly add new tests to a separate isolation
> > framework (called isolation2) which uses a completely different
> > syntax. It doesn't use isolationtester at all. So, I haven't had a use
> > case to add long options to isolationtester yet :)
>
> Is that other framework somehow more capable?
>
The ability to declare a step as blocking, as Melanie mentioned upthread
("&" prefix), makes it more capable. The tester, when encounters such a
step, makes sure that the command in that step is blocking and moves on to
run subsequent commands. The isolationtester, on the other hand, treats a
step as blocking only when the command waits on a lock. That seems
restrictive. E.g. what if a command waits on a latch, as part of a valid
interleaving of concurrent transactions? The isolation tester cannot
detect such a case and it will keep waiting and eventually fail the test
with a timeout.
As part of the fault injector patch set [1], I added a new "blocking"
keyword to isolation grammar so that a step can be declared as blocking.
See patch 0002-Add-syntax-to-declare-a-step-that-is-expected-to-block.
Asim
[1]
https://www.postgresql.org/message-id/CANXE4TdxdESX1jKw48xet-5GvBFVSq%3D4cgNeioTQff372KO45A%40mail.gmail.com
