You are viewing a plain text version of this content. The canonical link for it is here.
Posted to replication@couchdb.apache.org by Traun Leyden <tr...@gmail.com> on 2014/07/08 15:42:06 UTC
Question on replication process
I'm working on a 3rd party replication tool
<https://github.com/couchbaselabs/sg-replicate> in Go (currently Sync
Gateway compatible, but hopefully one day will be CouchDB compatible as
well), and I have a question on the replication process.
>From this replication algorithm description
<https://github.com/couchbaselabs/TouchDB-iOS/wiki/Replication-Algorithm>
and ladder diagram <http://cl.ly/image/1v013o210345>, if the
_rev_diffs response is empty, should it then skip steps 5 and 6 since there
is nothing to transfer and go straight to step 7? (using the last_seq
value that was returned from source db _changes feed in step 3)
Or should it skip step 7 too and do another _changes request?
I'm guessing it's the former, since it seems like it should be safe to
forward the checkpoint if the _rev_diffs response is empty. I haven't had
a chance to study the CouchDB replicator behavior yet, and thought someone
here might know off the top of their head.
Re: Question on replication process
Posted by Traun Leyden <tr...@gmail.com>.
Thanks!
> On Jul 8, 2014, at 6:46 AM, Dale Harvey <da...@arandomurl.com> wrote:
>
> If _revs_diff is empty, then yes 5 and 6 arent needed, as they mention they
> operate on the result of the revs_diff, if theres nothing to operate on
>
> You can skip 7 it just means you miss the optimisation and next time you do
> replication you will replicate redundant documents, checkpointing is purely
> an optmisation to reduce the number of redundant document replications
>
>
>> On 8 July 2014 14:42, Traun Leyden <tr...@gmail.com> wrote:
>>
>> I'm working on a 3rd party replication tool
>> <https://github.com/couchbaselabs/sg-replicate> in Go (currently Sync
>> Gateway compatible, but hopefully one day will be CouchDB compatible as
>> well), and I have a question on the replication process.
>>
>> From this replication algorithm description
>> <https://github.com/couchbaselabs/TouchDB-iOS/wiki/Replication-Algorithm>
>> and ladder diagram <http://cl.ly/image/1v013o210345>, if the
>> _rev_diffs response is empty, should it then skip steps 5 and 6 since there
>> is nothing to transfer and go straight to step 7? (using the last_seq
>> value that was returned from source db _changes feed in step 3)
>>
>> Or should it skip step 7 too and do another _changes request?
>>
>> I'm guessing it's the former, since it seems like it should be safe to
>> forward the checkpoint if the _rev_diffs response is empty. I haven't had
>> a chance to study the CouchDB replicator behavior yet, and thought someone
>> here might know off the top of their head.
>>
Re: Question on replication process
Posted by Dale Harvey <da...@arandomurl.com>.
If _revs_diff is empty, then yes 5 and 6 arent needed, as they mention they
operate on the result of the revs_diff, if theres nothing to operate on
You can skip 7 it just means you miss the optimisation and next time you do
replication you will replicate redundant documents, checkpointing is purely
an optmisation to reduce the number of redundant document replications
On 8 July 2014 14:42, Traun Leyden <tr...@gmail.com> wrote:
> I'm working on a 3rd party replication tool
> <https://github.com/couchbaselabs/sg-replicate> in Go (currently Sync
> Gateway compatible, but hopefully one day will be CouchDB compatible as
> well), and I have a question on the replication process.
>
> From this replication algorithm description
> <https://github.com/couchbaselabs/TouchDB-iOS/wiki/Replication-Algorithm>
> and ladder diagram <http://cl.ly/image/1v013o210345>, if the
> _rev_diffs response is empty, should it then skip steps 5 and 6 since there
> is nothing to transfer and go straight to step 7? (using the last_seq
> value that was returned from source db _changes feed in step 3)
>
> Or should it skip step 7 too and do another _changes request?
>
> I'm guessing it's the former, since it seems like it should be safe to
> forward the checkpoint if the _rev_diffs response is empty. I haven't had
> a chance to study the CouchDB replicator behavior yet, and thought someone
> here might know off the top of their head.
>