Re: [HACKERS] BUG #12070: hstore extension: hstore_to_json_loose produces invalid JSON

On 1/15/15 2:29 PM, Bruce Momjian wrote:
> On Tue, Jan 13, 2015 at 10:56:48AM -0500, Peter Eisentraut wrote:
>> On 11/30/14 11:45 AM, Tom Lane wrote:
>>> The "API break" isn't a big issue imo.  The net effect would be that eg
>>> hstore 9.3.6 wouldn't work against a 9.3.5 server.  We do that sort of
>>> thing *all the time* --- at least twice in the past year, according to
>>> a quick scan of the commit logs.  If you were changing or removing a
>>> function that third-party code might depend on, it'd be problematic,
>>> but an addition has no such risk.
>>
>> This sort of things is actually a bit of an annoyance, because it means
>> that for minor-version upgrades, you need to stop the server before
>> unpacking the new version, otherwise the old running server will try to
>> load the new hstore module and fail with a symbol lookup.  This can
>> increase the downtime significantly.
>>
>> Yes, we've done this before, and people have gotten bitten by it before.
> 
> Uh, do we ever support installing new binaries while the server is
> running?  I would hope not.

Effectively, we don't, but it's not unreasonable to expect it.  Check
how your operating system upgrades other server packages such as apache
or openssh.