couchdb-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Newson <>
Subject Re: Ready for 0.10 Release?
Date Mon, 17 Aug 2009 22:21:26 GMT
Do ensure you're using the latest patch (the 3rd one I've posted to
the jira ticket) as it uses short, random leaps between ids to prevent
monotonicity (the concern being that ids are too guessable). i noticed
no reduction in improvement with that over strictly monotonic
increments, but I'd appreciate a double check.


On Mon, Aug 17, 2009 at 11:14 PM, Robert
Dionne<> wrote:
> I've tested with this quite a bit now and it's much faster, in some cases 4
> times, but it is sensitive to both doc size and batch sizes so in some cases
> the speed may only be about 4-10% better. It's definitely much improved in
> the file footprint resulting in faster compaction also.
> It's interesting to note that a good amount of the speed improvement comes
> from a reduced number of calls to couch_util:to_hex because it's only used
> for the prefix and not for every id as it is in the new_uuid() case. In one
> of the hovercraft tests I ran this method was cut from 36% to 25% of the run
> time. You can see this using eprof. This method is also used in the rev
> stuff so it's a good candidate to look at for optimization.
> So I"m +1 on this, but I think the other devs ought to noodle it some to
> double check it conceptually, .ie. that we have global uniqueness and it
> won't present issues elsewhere. Definitely a nice hack.
> Cheers,
> Bob
> On Aug 17, 2009, at 5:18 PM, Robert Newson wrote:
>> I was hoping for more feedback on COUCHDB-465 as the alternate uuid
>> generation algorithm there seems to double insertion speed and reduces
>> the  need to run compaction.
>> B.
>> On Mon, Aug 17, 2009 at 10:10 PM, Noah Slater<> wrote:
>>> On Mon, Aug 17, 2009 at 01:58:56PM -0700, Chris Anderson wrote:
>>>> We've been sprinting extra hard for the last week to get a lot of new
>>>> power into CouchDB. I think we're close on Windows support, we just
>>>> got native view servers, we've added a bunch of robustness around
>>>> external process handling. There's a bunch more that's happened since
>>>> 0.9 but the point of this thread is not to list the changes, but to
>>>> see if there are patches that people feel should go in now, so they
>>>> make the 0.10 release. Hopefully this is a short list, as I think most
>>>> of the last-minute patching has been done, but if something's
>>>> important and easy, we shouldn't forget it.
>>> I don't want to be the damp squid[1] here, but for now, shipping with the
>>> recent
>>> changes for Windows, or at least claiming Windows support, will get my -0
>>> or -1
>>> vote depending on how convincing people's arguments get.
>>> I wasn't even aware that this had properly landed yet, and it would
>>> certainly
>>> need to be simmered in trunk for a few months, with a wide call for
>>> testing,
>>> before I would be happy shipping it.
>>> [1]
>>> Best,
>>> --
>>> Noah Slater,

View raw message