Re: pg_basebackup vs. Windows and tablespaces

On 05/11/2015 11:01 PM, Amit Kapila wrote:
>
>
> On Tue, May 12, 2015 at 5:50 AM, Andrew Dunstan <andrew(at)dunslane(dot)net
> <mailto:andrew(at)dunslane(dot)net>> wrote:
> >
> >
> > On 05/11/2015 02:02 AM, Amit Kapila wrote:
> >>
> >> On Sun, May 10, 2015 at 6:01 AM, Andrew Dunstan
> <andrew(at)dunslane(dot)net <mailto:andrew(at)dunslane(dot)net>
> <mailto:andrew(at)dunslane(dot)net <mailto:andrew(at)dunslane(dot)net>>> wrote:
> >> >
> >> >
> >> >
> >> > This generally looks good, but I have a couple of questions
> before I commit it.
> >> >
> >> > First, why is the new option for the BASE_BACKUP command of the
> Streaming Replication protcol "TAR"? It seems rather misleading.
> Shouldn't it be something like "TABLESPACEMAP"?
> >> >
> >>
> >> The reason to keep new option's name as TAR was that tablespace_map
> >> was generated for that format type, but I agree with you that something
> >> like "TABLESPACEMAP" suits better, so I have changed it to
> >> "TABLESPACE_MAP". Putting '_' in name makes it somewhat consistent
> >> with other names and filename it generates with this new option.
> >>
> >>
> >> > Second, these lines in xlog.c seem wrong:
> >> >
> >> > else if ((ch == '\n' || ch == '\r') && prev_ch == '\\')
> >> > str[i-1] = '\n';
> >> >
> >> > It looks to me like we should be putting ch in the string, not
> arbitrarily transforming \r into \n.
> >> >
> >>
> >> You are right, I have changed it as per your suggestion.
> >>
> >>
> >
> >
> > OK, I have cleaned this up a bit - I had already started so I didn't
> take your latest patch but instead applied relevant changes to my
> changeset. Here is my latest version.
> >
> > In testing I notice that now "pg_baseback -F t" leaves it completely
> up to the user on all platforms to create the relevant links in
> pg_tblspc/. It includes the tablespace_map file in base.tar, but
> that's really just informational.
> >
>
> Sorry, but I am not able to follow your point. User don't need to create
> the relevant links, they will get created during first startup (during
> recovery)
> from the backup. I have tested and it works both on Windows and Linux.
>
> Refer below code of patch in xlog.c
>
> +
> + /* read the tablespace_map file if present and create symlinks. */
> + if (read_tablespace_map(&tablespaces))
> + {
> ..
>
> > I think we need to add something to the pg_basebackup docs about
> that, at the very least (and it will also need to be a release note item.)
> >
>
> Currently, below lines in patch suggest that this file is required for
> recovery.
> Do you expect more information to be added?
>
> + These files are not merely for your information; their presence and
> + contents are critical to the proper operation of the system's
> recovery
> + process.
>

Yes, sorry, I had a moment of brain fade yesterday. However, I think
we're a bit under-documented on the pg_basebackup page, regarding both
tar mode and tablespace_map (which isn't even mentioned).

And there is this which I noticed this morning: the --tablespace-mapping
feature of pg_basebackup seems to be quite broken in --format=tar mode -
it simply has no effect AFAICT. I assume it was broken before, but we
should either fix it (possibly hard) or disallow the combination (which
would be a pity).

I'm going to go ahead and commit this in the state I have it now,
because for the most part these are preexisting deficiencies.

cheers

andrew