On Sun, Mar 26, 2017 at 3:34 AM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
> I wrote:
>> It doesn't seem to be a platform-specific problem: I can duplicate the
>> failure here by applying same settings mandrill uses, ie build with
>> -DRANDOMIZE_ALLOCATED_MEMORY and set force_parallel_mode = regress.
>
> Oh ... scratch that: you don't even need -DRANDOMIZE_ALLOCATED_MEMORY.
> For some reason I just assumed that any parallelism-related patch
> would have been tested with force_parallel_mode = regress. This one
> evidently was not.
>
> regards, tom lane
>
This is caused because trigger related functions are marked safe and
using global variables, hence when executed in parallel are giving
incorrect output. Attached patch fixes this. I have modified only
those functions that are showing incorrect behaviour in the regress
output and checked regression test with force_parallel_mode = regress
and all testcases are passing now.
This concerns me that should we be checking all the system defined
functions once again if they are actually parallel safe...?
--
Regards,
Rafia Sabih
EnterpriseDB: http://www.enterprisedb.com/