Horton

Re: Bugs in TOAST handling, OID assignment and redo recovery

Pavan Deolasee <pavan.deolasee@gmail.com>

View thread View on postgr.es
From: Pavan Deolasee <pavan.deolasee@gmail.com>
To: Peter Eisentraut <peter.eisentraut@2ndquadrant.com>
Cc: pgsql-hackers <pgsql-hackers@postgresql.org>
Date: 2018-04-18T06:37:52Z
Lists: pgsql-hackers 
On Thu, Apr 12, 2018 at 5:53 AM, Peter Eisentraut <
peter.eisentraut@2ndquadrant.com> wrote:

> On 4/10/18 06:29, Pavan Deolasee wrote:
> > One of our 2ndQuadrant support customers recently reported a sudden rush
> > of TOAST errors post a crash recovery, nearly causing an outage. Most
> > errors read like this:
> >
> > ERROR: unexpected chunk number 0 (expected 1) for toast value nnnn
>
> While researching this, I found that the terminology in this code is
> quite inconsistent.  It talks about chunks ids, chunk indexes, chunk
> numbers, etc. seemingly interchangeably.  The above error is actually
> about the chunk_seq, not about the chunk_id, as one might think.
>
> The attached patch is my attempt to clean this up a bit.  Thoughts?
>

While I agree that we should clean it up, I wonder if changing error text
would be a good idea. These errors are being reported by a very long time
and if we change the text, we might forget the knowledge about the past
reports.

Also, "toast value" is same as "chunk_id". Should we clean up something
there too? "chunk_seq number" -- should that be just "chunk_seq"?


Thanks,
Pavan

-- 
 Pavan Deolasee                   http://www.2ndQuadrant.com/
 PostgreSQL Development, 24x7 Support, Training & Services

Commits

  1. Ignore nextOid when replaying an ONLINE checkpoint.

  2. Do not select new object OIDs that match recently-dead entries.