Re: Problems with pg_upgrade after change of unix user running db.

On Mon, Nov 30, 2015 at 08:08:50AM +0000, Benedikt Grundmann wrote:
>
>
> On Sat, Nov 28, 2015 at 2:39 AM, Adrian Klaver <adrian.klaver@aklaver.com>
> wrote:
>
>     On 11/27/2015 06:07 PM, Tom Lane wrote:
>     > Adrian Klaver <adrian.klaver@aklaver.com> writes:
>     >> On 11/27/2015 08:15 AM, Bruce Momjian wrote:
>     >>> My guess is you are sharing the constraint name "seqno_not_null" with
>     >>> multiple tables.  I think you are going to have to dig into the system
>     >>> tables to see where that is referenced and fix it.
>     >
>     >> In the post below the OP shows the tables involved(they where
>     inherited):
>     >> http://www.postgresql.org/message-id/
>     CADbMkNM_y9ewdaWdQ_8DJ1mUC0Z_FGwTyAD2RwCHgExj2jvOHQ@mail.gmail.com
>     >
>     > Inherited eh?  Maybe related to 074c5cfbf.
>
>
> I forgot to mention this earlier. This cluster is running 9.2.6 and I'm
> attempting to upgrade to the latest 9.4.5

Well, 9.4.5 we released on October 8, 2015, and the commit mentioned happened
on November 20, 2015, so that fix is not in 9.4.5:

    commit 074c5cfbfb4923158be9ccdb77420d6522d77538
    Author: Tom Lane <tgl@sss.pgh.pa.us>
    Date:   Fri Nov 20 14:55:28 2015 -0500

    Fix handling of inherited check constraints in ALTER COLUMN TYPE (again).

    The previous way of reconstructing check constraints was to do a separate
    "ALTER TABLE ONLY tab ADD CONSTRAINT" for each table in an inheritance
    hierarchy.  However, that way has no hope of reconstructing the check
    constraints' own inheritance properties correctly, as pointed out in
    bug #13779 from Jan Dirk Zijlstra.  What we should do instead is to do
    a regular "ALTER TABLE", allowing recursion, at the topmost table that
    has a particular constraint, and then suppress the work queue entries
    for inherited instances of the constraint.

    Annoyingly, we'd tried to fix this behavior before, in commit 5ed6546cf,
    but we failed to notice that it wasn't reconstructing the pg_constraint
    field values correctly.

    As long as I'm touching pg_get_constraintdef_worker anyway, tweak it to
    always schema-qualify the target table name; this seems like useful backup
    to the protections installed by commit 5f173040.

    In HEAD/9.5, get rid of get_constraint_relation_oids, which is now unused.
    (I could alternatively have modified it to also return conislocal, but that
    seemed like a pretty single-purpose API, so let's not pretend it has some
    other use.)  It's unused in the back branches as well, but I left it in
    place just in case some third-party code has decided to use it.

    In HEAD/9.5, also rename pg_get_constraintdef_string to
    pg_get_constraintdef_command, as the previous name did nothing to explain
    what that entry point did differently from others (and its comment was
    equally useless).  Again, that change doesn't seem like material for
    back-patching.

    I did a bit of re-pgindenting in tablecmds.c in HEAD/9.5, as well.

    Otherwise, back-patch to all supported branches.

Are you able to compile from 9.4 git head and test that?  It seems
dumping inheriting contraints from parents has not worked properly for
some time.