>>> On Fri, Oct 12, 2007 at 12:37 PM, in message
> Tom Lane wrote:
>> "Joshua D. Drake" <[EMAIL PROTECTED]> writes:
>>> With respect to you Kevin, your managers should wait. You don't
>>> install .0 releases of "any" software into production without "months"
>>> of testing. At which point, normally a .1 release has come out anyway.
All generalities are false.
Everyone needs to assess the risks and benefits for their own
environment, and figure out how close the can comfortably be to the
bleeding edge on any new technology.
We have been able to exercise not only .0 but RC and beta releases in
production because we have central servers holding consolidated
copies of the various distributed originals; with multiple copies of
these databases.  So we have redundancy in two dimensions, not to
mention backups of the redundant databases and two parallel backup
strategies for the original data.  And we synchronize the originals
to the copies during any slack time, and investigate any
discrepancies.  If the release seems stable enough, we'll put one copy
of the live redundant system at risk on beta or RC.
Is this testing or production?  I guess you could argue the semantics
either way.  Of course we do some load tests by replaying the
production HTTP request stream through test renderers; but it's being
fed 24/7 from a production environment, and we've been known to use
it for ad hoc queries.  I think we've even load-shifted the web site
over to it briefly to perform maintenance on the other servers.
>> How exactly do you expect the software to get from a .0 to a .1 release,
>> or to have addressed the bugs that might bite you when it does get to .1,
>> if you aren't helping to test it?
> In most environments I've seen, developer and QA systems don't hesitate
> to move to .0 releases (or even beta).   I agree with Joshua that it's
> nerve wracking to move _production_ systems to .0 releases from most
> software vendors.
My philosophy is that the final QA environment should be as close to
the production environment as can be arranged, but that difference in
the development and initial test environments contribute to

---------------------------(end of broadcast)---------------------------
TIP 9: In versions below 8.0, the planner will ignore your desire to
       choose an index scan if your joining column's datatypes do not

Reply via email to