Horton

Re: Potential G2-item cycles under serializable isolation

Thomas Munro <thomas.munro@gmail.com>

View thread View on postgr.es
From: Thomas Munro <thomas.munro@gmail.com>
To: Kyle Kingsbury <aphyr@jepsen.io>
Cc: PostgreSQL mailing lists <pgsql-bugs@lists.postgresql.org>
Date: 2020-06-02T23:13:02Z
Lists: pgsql-bugs 
On Wed, Jun 3, 2020 at 4:19 AM Kyle Kingsbury <aphyr@jepsen.io> wrote:
> https://github.com/jepsen-io/jepsen/blob/f47eb25ab32529a7b66f1dfdd3b5ef2fc84ed778/stolon/src/jepsen/stolon/append.clj#L31-L108

I'm looking into this, but just by the way, you said:

; OK, so first worrying thing: why can this throw duplicate key errors if
; it's executed with "if not exists"?
(try
  (j/execute! conn
    [(str "create table if not exists " (table-name i)
    ....

That's (unfortunately) a known problem under concurrency.  It'd be
very nice to fix that, but it's an independent problem relating to DDL
(not just tables, and not just IF EXISTS DDL, but anything modifying
catalogues can race in this way and miss out on "nice" error messages
or the IF EXISTS no-op).  Here's a good short summary:

https://www.postgresql.org/message-id/CA%2BTgmoZAdYVtwBfp1FL2sMZbiHCWT4UPrzRLNnX1Nb30Ku3-gg%40mail.gmail.com

Commits

  1. Doc: Add references for SI and SSI.

  2. Improve comments for [Heap]CheckForSerializableConflictOut().

  3. Avoid update conflict out serialization anomalies.

  4. Implement genuine serializable isolation level.