Re: Invalid pointer access in logical decoding after error

On Thu, Oct 9, 2025, at 10:40 AM, vignesh C wrote:
> On Thu, 9 Oct 2025 at 00:16, Masahiko Sawada <sawada(dot)mshk(at)gmail(dot)com> wrote:
>>
>> One thing we might want to consider is for v14 and v13. They don't
>> have this bug since the entry_ctx was introduced in v15, but it's
>> still true for them that RelationSyncCache is not cleaned up in error
>> cases if pgoutput is used via SQL API. For example, if
>> RelationSyncCache hash table gets corrupted for some reason, logical
>> decoding could repeat an error until logical decoding completes
>> successfully and its shutdown callback is called. Also, it might be a
>> good idea in general to ensure cleaning up the hash table after use.
>
> Agreed, let's backpatch to PG13. Should we also add a test case in the
> master branch, given that this issue has been around for a while?
>

I'm wondering if it is a good idea because the bug doesn't manifest in v13 and
v14. At least the v13 has its final minor version in less than a month and EOL.
I would have caution when applying fixes to the latest minor version of a
stable branch; there won't be a chance to fix the fix in the next minor
release. Furthermore, in these back branches, the patch doesn't fix a known
issue. I wouldn't bother with these back branches. For v14, if, in a couple of
months, we have some reports that justify the backpatch, fine.

--
Euler Taveira
EDB https://www.enterprisedb.com/