Re: PATCH: Warn users about tablespace abuse data loss risk

2014-02-12 15:16 GMT+09:00 Ian Lawrence Barwick <barwick(at)gmail(dot)com>:
> 2014-02-12 14:06 GMT+09:00 David Johnston <polobo(at)yahoo(dot)com>:
>> Ian Lawrence Barwick wrote
>>> 2014-02-12 12:52 GMT+09:00 Craig Ringer <
>>
>>> craig@
>>
>>> >:
>>>> Hi all
>>>>
>>>> I've just seen another case of data loss due to misuse of /
>>>> misunderstanding of tablespaces:
>>>>
>>>> http://dba.stackexchange.com/questions/58704/how-do-i-access-a-old-saved-tablespace-after-reinstalling-postgres
>>>>
>>>> and it's prompted me to write some docs amendments to make it more
>>>> obvious that *you shouldn't do that*.
>>>>
>>>> Not that it'll stop people, but it'll at least mean they can't say we
>>>> didn't warn them.
>>>>
>>>> This is actually quite important, because many users are used to MySQL's
>>>> MyISAM, where each table contains its own metadata and is readable by
>>>> simply copying the table into a different MySQL install's data
>>>> directory. It doesn't even have to be the same version! Users are
>>>> clearly surprised that PostgreSQL tablespaces don't have the same
>>>> properties.
>>>>
>>>> Thoughts?
>>>
>>> People still use MyISAM!?
>>>
>>> I had a similar issue pop up at work a while back, having something
>>> explicit to point to is definitely a good idea.
>>>
>>> Suggestion for the first paragraph of the patch (sorry I can't provide it
>>> in
>>> patch form right now):
>>>
>>> Even if they are located outside the main PostgreSQL data directory,
>>> tablespaces
>>> are an integral part of the database cluster and
>>> <emphasis>
>>> cannot
>>> </emphasis>
>>> be
>>> treated as an autonomous collection of data files. They rely on
>>> metadata contained
>>> in the main data directory, without which they are useless. In
>>> particular, tablespaces
>>> cannot be reattached to a different database cluster, and backing up
>>> individual
>>> tablespaces makes no sense as a backup/redundancy method. Similarly,
>>> if you lose a
>>> tablespace (file deletion, disk failure, etc) the main database may
>>> become unreadable
>>> or fail to start.
>>>
>
>> While providing additional warnings is good and necessary it may also help
>> to be more descriptive as to in what situations tablespaces are appropriate
>> and/or necessary so that people leave with a better understanding of why the
>> feature exists and not just trying to know what not to use it for. It goes
>> against the more prescriptive tone of the documentation generally but both
>> approaches work well together to tackle the knowledge/understanding gap some
>> users seem to have.
>
> The warning would appear on this page:
>
> http://www.postgresql.org/docs/current/static/manage-ag-tablespaces.html
>
> which describes what tablespaces *can* do, but unless you're familiar with the
> structure of the PostgreSQL data directories, it's not obvious what you *can't*
> do. I recall reading a blog post a while back about tablespaces being "archived"
> to the cloud with disastrous results, and a quick search pulls up
> stuff like this:
>
> http://stackoverflow.com/questions/3534415/moving-postgres-tablespaces-and-tables-across-ec2-instance
>
> so it's definitely not a niche issue. Something "official" to link to
> would be very useful in this kind of situation. That doesn't preclude the general
> documentation being improved of course.

And taking a look at the page in question I see this prominent example:

CREATE TABLESPACE fastspace LOCATION '/mnt/sda1/postgresql/data';

As '/mnt' is usually a temporary mount point for detachable media and
all that [*]
maybe it's not the best impression to give for a suitable tablespace location.

[*] http://www.pathname.com/fhs/pub/fhs-2.3.html#MNTMOUNTPOINTFORATEMPORARILYMOUNT
"This directory is provided so that the system administrator may
temporarily mount a
filesystem as needed. "

Ian Barwick