Re: pg_rewind and xlogtemp files

On Wed, Jun 17, 2015 at 9:07 PM, Fujii Masao <masao(dot)fujii(at)gmail(dot)com> wrote:
> On Wed, Jun 17, 2015 at 4:57 PM, Vladimir Borodin <root(at)simply(dot)name> wrote:
>>
>> 17 июня 2015 г., в 9:48, Michael Paquier <michael(dot)paquier(at)gmail(dot)com>
>> написал(а):
>>
>> On Wed, Jun 17, 2015 at 3:17 PM, Michael Paquier
>> <michael(dot)paquier(at)gmail(dot)com> wrote:
>>
>> As pointed by dev1ant on the original bug report, process_remote_file
>> should ignore files named as pg_xlog/xlogtemp.*, and I think that this
>> is the right thing to do. Any objections for a patch that at the same
>> time makes "xlogtemp." a define declaration in xlog_internal.h?
>>
>>
>> Declaration seems to be the right thing.
>>
>> Another problem I’ve caught twice already in the same test:
>>
>> error reading xlog record: record with zero length at 0/78000090
>> unexpected result while fetching remote files: ERROR: could not open file
>> "base/13003/t6_2424967" for reading: No such file or directory
>> The servers diverged at WAL position 0/76BADD50 on timeline 303.
>> Rewinding from Last common checkpoint at 0/7651F870 on timeline 303
>>
>> I don’t know if this problem could be solved the same way (by skipping such
>> files)… Should I start a new thread for that?
>
> That's the file of the temporary table, so there is no need to copy it
> from the source server. pg_rewind can safely skip such file, I think.

Yes. It is actually recommended to copy them manually if needed from
the archive (per se the docs).

> But even if we make pg_rewind skip such file, we would still get the
> similar problem. You can see the problem that I reported in other thread.
> In order to address this type of problem completely, we would need
> to apply the fix that is been discussed in that thread.
> http://www.postgresql.org/message-id/CAHGQGwEdsNgeNZo+GyrzZtjW_TkC=XC6XxrjuAZ7=X_cj1aHHg@mail.gmail.com

There are two things to take into account here in my opinion:
1) Ignoring files that should not be added into the filemap, like
postmaster.pid, xlogtemp, etc.
2) bypass the files that can be added in the file map, for example a
relation file or a fsm file, and prevent erroring out if they are
missing.

> BTW, even pg_basebackup doesn't skip the file of temporary table.
> But maybe we should change this, too.
>
> Also pg_rewind doesn't skip the files that pg_basebackup does. ISTM
> that basically pg_rewind can safely skip any files that pg_basebackup does.
> So probably we need to reconsider which file to make pg_rewind skip.

pg_rewind and basebackup.c are beginning to share many things in this
area, perhaps we should consider a common routine in let's say
libpqcommon to define if a file can be safely skipped depending on its
path name in PGDATA.
--
Michael