On Fri, Sep 2, 2011 at 10:34 AM, Kevin Grittner
<Kevin.Grittner@wicourts.gov> wrote:
> Dimitri Fontaine <dimitri@2ndQuadrant.fr> wrote:
>> Robert Haas <robertmhaas@gmail.com> writes:
>>>> I'm also wondering if providing some shell script examples of a
>>>> fault-tolerant script to handle archiving would be useful.
>>>
>>> I think it would.
>>
>> My usual advice is to avoid having to write one if possible,
>> because it's more complex than it looks. What about recommending
>> existing solutions, such as walmgr from Skytools?
>>
>> Even better, what about including a default archiving tool, that
>> could be either another script in bin/ or rather an internal
>> command. The default would accept a location as argument, for
>> simple needs you mount a remote filesystem and there you go. If
>> you need something more complex, you still can provide it
>> yourself.
>
> In a green field I might argue for having an archvie_directory GUC
> instead of archive_command. As it stands, it might be a really good
> idea to provide a pg_archiveto executable which takes as arguments a
> directory path and the arguments passed to the archive script. With
> a little extra effort, the executable could check for some file
> which would specify what host and path should be writing archives
> there, to avoid problems with copied database directories
> accidentally writing to the same location as the source.
>
> Such an executable seems like minimal effort compared to the
> problems it would solve.
>
> If there's an existing tool with appropriate licensing which is
> sufficiently portable and reliable, all the better -- let's ship it
> and use that for our example archive_command.
Another thought I have here is to wonder whether we should change
something on the server side so that we don't NEED such a complicated
archive_command. I mean, copying a file to a directory somewhere is
not fundamentally a complex operation. Nor is using ssh to copy it to
another machine. The fact that archive_commands need to be so complex
seems like a usability defect. The consensus seems to be that just
using something like 'cp' for your archive command won't work out
well, but maybe instead of shipping a more complicated script we
should be trying to eliminate (or at least reduce) the need for a more
complicated script.
--
Robert Haas
EnterpriseDB: http://www.enterprisedb.com
The Enterprise PostgreSQL Company