Re: [HACKERS] Broken hint bits (freeze)

On Fri, Jun 23, 2017 at 8:47 PM, Sergey Burladyan <eshkinkot@gmail.com> wrote:
> Bruce Momjian <bruce@momjian.us> writes:
>
>> On Wed, Jun 21, 2017 at 07:49:21PM +0530, Amit Kapila wrote:
>> > On Tue, Jun 20, 2017 at 7:24 PM, Amit Kapila <amit.kapila16@gmail.com> wrote:
>> > > Hmm.  I think we need something that works with lesser effort because
>> > > not all users will be as knowledgeable as you are, so if they make any
>> > > mistakes in copying the file manually, it can lead to problems.  How
>> > > about issuing a notification (XLogArchiveNotifySeg) in shutdown
>> > > checkpoint if archiving is enabled?
>> > >
>> >
>> > I have thought more about the above solution and it seems risky to
>> > notify archiver for incomplete WAL segments (which will be possible in
>> > this case as there is no guarantee that Checkpoint record will fill
>> > the segment).  So, it seems to me we should update the document unless
>> > you or someone has some solution to this problem.
>
>> As far as I know this is the only remaining open issue.  Sergey, please
>> verify.  I appreciate the work everyone has done to improve this, and
>> all the existing fixes have been pushed to all supported branches.  :-)
>
> Yes, thank you all for your help!
>
> Yes, this is last issue with checkpoint that I know, how to ensure that
> standby sync all shared buffers into disk on it shutdown.
>

I think if we have a command like Alter System Flush Shared Buffers,
then it would have been helpful in what you need here.  You could have
run it before shutdown.

> I thinking about enforce restartpoint on shutdown, like:
> src/backend/access/transam/xlog.c
> -   8639     if (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
> -   8640         XLByteLE(lastCheckPoint.redo, ControlFile->checkPointCopy.redo))
> -   8641     {
> +   8639     if ( !(flags & CHECKPOINT_IS_SHUTDOWN) && (XLogRecPtrIsInvalid(lastCheckPointRecPtr) ||
> +   8640         XLByteLE(lastCheckPoint.redo, ControlFile->checkPointCopy.redo) )
> +   8641     {
>
> But I still not read source and not sure about this solution.
>

It might serve your purpose, but I think it will not be safe to
perform restartpoint always at shutdown.  It will delete the WAL files
which should be deleted only after the actual checkpoint record is
received from the master.

>
> PS:
> I successfully upgraded last night from 9.2 to 9.4 and find other issue :-)
>
> It is about hash index and promote:
> 1. create master
> 2. create standby from it
> 3. create unlogged table and hash index like:
>  create unlogged table test (id int primary key, v text);
>  create index on test using hash (id);
> 3. stop master
> 4. promote standby
>
> now, if you try to upgrade this new promoted master pg_upgrade will stop
> on this hash index:
> error while creating link for relation "public.test_id_idx" ("s/9.2/base/16384/16393" to "m/9.4/base/16422/16393"):
Nosuch file or directory
 
> Failure, exiting
>

I am not sure if this is a problem because in the version you are
trying hash indexes are not WAL-logged and the creation of same will
not be replicated on standby, so the error seems to be expected.

-- 
With Regards,
Amit Kapila.
EnterpriseDB: http://www.enterprisedb.com