References: <AB5978B2-1772-4FEE-A245-74C91704ECB0@amazon.com>
 <87ftjifoql.fsf@ars-thinkpad>
 <20191024213157.7pm6niybfxgpvmgg@alap3.anarazel.de>
 <87eez1fh48.fsf@ars-thinkpad>
 <CAA4eK1L=MDbmGu5-+BmY7Svc07jr+ZabiH8C_qo3RSc8pgUpDQ@mail.gmail.com>
 <8736bs81sx.fsf@ars-thinkpad>
 <CAA4eK1LdNmrib1jub8b=KvYUrzXW0VT4P3MVPMyiMfMY3K64dA@mail.gmail.com>
 <871rrb942q.fsf@ars-thinkpad>
 <CAA4eK1LYzrZ_+8VhD_N_dsQwjxA9t+AyGKT-Wjnc8S7jCwAcBw@mail.gmail.com>
 <87zhdx76d5.fsf@ars-thinkpad>
 <CAA4eK1Jdh0zab=+D91MkFPbevzyjFCZsMwRsQxJh4F9+m_vCRA@mail.gmail.com>
 <CAA4eK1JnSKkNdgLAHWY+YCE_3Li454So_thyTyNJa_G3hSVscA@mail.gmail.com>
 <8736bjoiax.fsf@ars-thinkpad>
 <CAA4eK1Kcsib6UG7zFPrL-h2fvnByfKWxq8Xvzg=2hxUebYwt=g@mail.gmail.com>
 <87wo8ulhjl.fsf@ars-thinkpad>
 <CAA4eK1+zGWCTsPqpL0fON6xW8BWKp_JtHr72hio3RDLJnv0G-A@mail.gmail.com>
 <87lfp9u4kp.fsf@ars-thinkpad>
 <CAA4eK1L_i7UDLxC6apa=bt+KxVPyEN0L4r1ig-oiOXksEH3jpQ@mail.gmail.com>
User-agent: mu4e 1.2.0; emacs 26.0.50
From: Arseny Sher <a.sher@postgrespro.ru>
To: Amit Kapila <amit.kapila16@gmail.com>
Cc: Arseny Sher <a.sher@postgrespro.ru>, Andres Freund <andres@anarazel.de>,
 "Hsu\, John" <hsuchen@amazon.com>,
 "pgsql-bugs\@lists.postgresql.org" <pgsql-bugs@lists.postgresql.org>
Subject: Re: ERROR: subtransaction logged without previous top-level txn
 record
In-reply-to: 
 <CAA4eK1L_i7UDLxC6apa=bt+KxVPyEN0L4r1ig-oiOXksEH3jpQ@mail.gmail.com>
Date: Tue, 11 Feb 2020 15:06:02 +0300
Message-ID: <87tv3x72hh.fsf@ars-thinkpad>
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="=-=-="
Precedence: bulk

--=-=-=
Content-Type: text/plain


Amit Kapila <amit.kapila16@gmail.com> writes:

>> That's weird, it reliably fails with expected error for me. There are
>> already two s2_checkpoint's: first establishes potential (broken)
>> restart_lsn (serializes snapshot after first xl_xact_assignment of s0
>> xact, but before first record of s1 xact), the second ensures
>> s2_get_changes directly following it will actually advance the slot,
>>
>
> In my case, s2_get_changes doesn't seem to be advancing the restart
> lsn because when it processed running_xact by s2_checkpoint, the slots
> confirm flush location (slot->data.confirmed_flush) was behind it.  As
> confirmed_flush was behind running_xact of s2_checkpoint, it couldn't
> update slot->candidate_restart_lsn (in function
> LogicalIncreaseRestartDecodingForSlot).  I think the confirmed_flush
> location will only be updated at the end of get_changes.  This is the
> reason I need extra get_changes call to generate an error.
>
> I will think and investigate this more, but thought of sharing the
> current situation with you.  There is something different going on in
> my system or maybe the nature of test is like that.

Ah, I think I know what's happening -- you have one more
xl_running_xacts which catches the advancement -- similar issue is
explained in the comment in oldest_xmin.spec.

Try attached.


--=-=-=
Content-Type: text/x-diff
Content-Disposition: inline;
 filename=0002-Stop-demanding-that-top-xact-must-be-seen-before--v4.patch