Re: pg_basebackup for streaming base backups

On Tue, Jan 18, 2011 at 03:14, Fujii Masao <masao(dot)fujii(at)gmail(dot)com> wrote:
> On Mon, Jan 17, 2011 at 10:50 PM, Magnus Hagander <magnus(at)hagander(dot)net> wrote:
>>> +       printf(_("  -D, --pgdata=directory   receive base backup into directory\n"));
>>> +       printf(_("  -T, --tardir=directory    receive base backup into tar files\n"
>>> +                        "                            stored in specified directory\n"));
>>> +       printf(_("  -Z, --compress=0-9        compress tar output\n"));
>>> +       printf(_("  -l, --label=label         set backup label\n"));
>>> +       printf(_("  -p, --progress            show progress information\n"));
>>> +       printf(_("  -v, --verbose             output verbose messages\n"));
>>>
>>> Can we list those options in alphabetical order as other tools do?
>>
>> Certainly. But it makes more sense to have -D and -T next to each
>> other, I think - they'd end up way elsewhere. Perhaps we need a group
>> taht says "target"?
>
> I agree with you if we end up choosing -D and -T for a target rather
> than pg_dump-like options I proposed.

Yeah. What do others think between tohse two options? -D/-T followed
by directory, or -D <dir> and -F<format>?

>> Requiring PQfinish() might be more reasonable since it will give you a
>> log on the server if you don't, but I'm not convinced that's necessary
>> either?
>
> At least it's required for each password-retry. Otherwise, previous
> connections remain during backup. You can see this problem easily

Oh yeah, I've put that one in my git branch already.

> by repeating the password input in pg_basebackup.
>
>    LOG:  could not send data to client: Connection reset by peer
>    LOG:  could not send data to client: Broken pipe
>    FATAL:  base backup could not send data, aborting backup
>
> As you said, if PQfinish is not called at exit(1), the above messages
> would be output. Those messages look ugly and should be
> suppressed whenever we *can*. Also I believe other tools would
> do that.

Yeah, agreed. I'll add that, shouldn't be too hard.

>>> +       keywords[2] = "fallback_application_name";
>>> +       values[2] = "pg_basebackup";
>>>
>>> Using the progname variable seems better rather than the fixed word
>>> "pg_basebackup".
>>
>> I don't think so - that turns into argv[0], which means that if you
>> use the full path of the program (/usr/local/pgsql/bin/pg_basebackup
>> for example), the entire path goes into fallback_application_name -
>> not just the program name.
>
> No. get_progname extracts the actual name.

Hmm. I see it does. I wonder what I did to make that not work.

Then I agree with the change :-)

--
 Magnus Hagander
 Me: http://www.hagander.net/
 Work: http://www.redpill-linpro.com/