hi,
I'm new to Riak and learning it rapidly. Other than Riak, I also tried
MongoDB, CouchDB, and Voldemort (and also Cassandra ) briefly. Riak is quite
suitable for my needs. Thanks for releasing it as an open source project.
I got a number of questions but let me ask those about the PBC API first
only. Just some backgrounds: I am implementing my own Java ProtoBuf client.
I do it for several reasons: Kresten claims his Java PB client performs ~10X
better than the Java REST client, PB is preferred to HTTP/REST "officially"
for Erlang development, and I want to take a different approach to use PB by
using Java NIO/Netty instead of making direct socket connection.
My questions are as follows:
1. For put and delete operations, the 'w' and 'dw' parameters are defined
as integer, but in the documentation, it says "possible values include
'default', 'one', 'quorum', 'all', or any integer <= N". so what value
represents 'default', 'quorum', and 'all'? does 0 means default? how about
-1?
2. REST API allows "If-None-Match, If-Match, If-Modified-Since, and
If-Unmodified-Since" checking. PBC API does not have equivalent, right? I
read that for Erlang, PB is the preferred client. But it doesn't provide all
features that REST API provide?
It seems to me the REST and BPC APIs are not very consistent atm. for
example, REST API doesn't provide a way to list buckets (Issue 78); other
than those "If-None-Match .." criterias, the document also mentioned a
number of missing features in BPC API and suggested us to use REST API for
those cases.
3. I want to clarify the expected response for a "Store a new or existing
object with a key" cases
- From the doc, "Riak can allow the last update to automatically "win" or
Riak can return both versions of the object to the client.", I suppose it
refer to the "allow_multi" bucket parameter. I watched the
Wriaki video and
roughly know how allow_multi=true works in keeping versions.
- for a bucket with allow_multi=false, when an update is executed:
- if vclock is not provided, Riak won't know I try to update an
existing object so any update shall always be successful.
- if supplied vclock doesn't match the the vlock value of the
current object, would it return error(RpbErrorResp)? or just
"allow the last
update to automatically win"? or would it return the sibling like
allow_multi=true?
The following sequence is for illustrating the scenario of the
above question:
1) client A (or whoever) put a new k1,v1, with vclock c1
2) client A and B both read the entry with c1
3) client A updates the entry to k1,v2, supplied c1, the vclock is
updated to c2
4) client B tries to update k1, supplied v3 and c1. Because the
vclock is updated to c2 in step 3, i expect it to return an
error, or at
least there should an easy way for to know about conflict.
4. I wonder what interface will be added to the PBC API in your
roadmap. there is essential no administrative interface (except set bucket
properties) or any SPI. Take Voldemort as an example, they provides an
AdminClient interface (
http://project-voldemort.com/javadoc/client/index.html?voldemort/client/protocol/admin/AdminClient.html)
and they also have API to get the node list for a key.
(DefaultStoreClient.getResponsibleNodes() ) Riak doesn't provide these kind
of interface.In Erlang console, I could get the virtual node list in the
ring with riak_core_ring_manager:get_my_ring(), could this kind of
information be exposed as an admin client interface?
There are two related issues from my search:
https://issues.basho.com/show_bug.cgi?id=271
https://issues.basho.com/show_bug.cgi?id=421
For Issue 421, does HTTP APIs refers to the the admin interface that I
mentioned above?
Riak provides a map reduce framework/feature, but I don't really prefer
to write complex mapreduce job in javascript or to learn Erlang. If there
are SPI for resolving the physical node for keys, I may use an external map
reduce framework like GridGain (or Hadoop) and run map-reduce process the
process data in Riak with local data affinity.
Please don't get me wrong as I like Riak very much so far. I am writing to
clarify things rather than ranting, ;-) as I'm new and probably may
understand the things incorrectly . thank you very much.
regards,
mingfai
_______________________________________________
riak-users mailing list
riak-users@lists.basho.com
http://lists.basho.com/mailman/listinfo/riak-users_lists.basho.com
