couchdb-user mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Robert Newson <>
Subject Re: filters for _changes dont seem to work
Date Sun, 29 Nov 2009 23:25:34 GMT
I've been unable to reproduce the first part of that (curse these fast
connections!) but my point was to confirm that gaps are definitely
part of the API and they exist for a good reason (gaps are not bugs).

Given that the _changes feed is used for replication and keeping
external indexers in sync (obviously my concern here is for
couchdb-lucene), I thought it important to clarify the kind of gap
that can occur and hoped to convey that they are not a problem at all
for reasonable uses of _changes.

For my project, I certainly would prefer to not see a flurry of
changes to the same document, since I would spend time applying a
change that was then immediately obsoleted. This is also the reason
that a non-continuous call to _changes doesn't emit every revision of
every document; it's unnecessary for replication or synchronization.
Of course, those older revisions can be removed at any time by
compaction, so it's not like they could ever be guaranteed.


On Sun, Nov 29, 2009 at 11:02 PM, Chris Anderson <> wrote:
> On Sun, Nov 29, 2009 at 2:33 PM, Robert Newson <> wrote:
>> I'm not sure the API is quite as you say. When using feed=continuous
>> you do see all the changes, since it's just being passed to an event
>> listener of some kind.
> I may be splitting hairs here, but I think even with continuous
> changes you may not see every single update to a particular doc. For
> instance if you are on a very slow connection and someone is updating
> the doc from a very fast connection. It's just important that people
> understand these guarantees.
>> Where you do, definitely, see 'gaps' is when
>> you use _changes retrospectively.
> Changes is just following the by_sequence index, so these gaps are
> normal. The point I'm making is that even with continuous changes you
> are following the same by_sequence index, so the guarantees are the
> same.
> Chris
> --
> Chris Anderson

View raw message