Online Presentation: http://slid.es/maximsokhatsky/kvs
- Polymorphic Tuples
- Managing Linked-Lists
- Various Backends Support: Mnesia, Riak, KAI, Redis, MongoDB
- Sequential Consistency via Feed Server
- Basic Schema for Social Sites and Accounting
- Extendable Schema
- Supports Secondary Indexes for KAI, Mnesia, Riak and MongoDB
- Change Backends on-the-fly
- Supports Multiple backends at the same time
- Xen Ready
In rebar.config:
{kvs, ".*", {git, "git://github.com/synrc/kvs", "HEAD"}}Redis also need to add:
{eredis, ".*", {git, "git://github.com/wooga/eredis", {tag, "v1.0.6"} }}MongoDB also need to add:
{mongodb, ".*", {git, "git://github.com/comtihon/mongodb-erlang", {tag, "master"} }},
{poolboy, ".*", {git, "git://github.com/devinus/poolboy", {tag, "master"} }}In the above example poolboy is optional. MongoDB and Poolboy config example:
{kvs, [
{mongo, [
{connection, [{database,<<"kvs">>}]},
{pool, [{size,10},{max_overflow,20}]}
]}
]}To disable poolboy exclude {pool, ...} from your sys.config. More information on the configuring MongoDB and Poolboy can be found here: https://github.com/comtihon/mongodb-erlang, https://github.com/devinus/poolboy.
We have built with KVS a number of applications and came up with schema samples. We grouped schemas by three category. KVS hides database access behind backend drivers and provides high-level rich API to stored and extend following data:
- Core — Acl, Users, Subscriptions, Feeds, Entries, Comments
- Banking — Account, Customer, Transaction, Item, Currency, Program, Card, Cashback
- Social — Group, Meeting, Payment, Product, Purchase
This Framework provides also a feed application for sequential consistency and cr application for chain replication database on top of kvs. All write requests with given object key will be handled by single processes so you may not worry about concurrent changes of user feed tops.
All write operations that are made to data with secondary indexes, i.e. not like linked lists could be potentially handled without feed_server. But some KV storages are not supporting secondary indexes so use these backends carefully.
Currently kvs includes following store backends:
- Mnesia
- Riak
- KAI
- Filesystem
- Redis
- MongoDB
First of all you need to tune your backend in the kvs application:
{kvs, [{dba,store_mnesia}]},Try to check it:
1> kvs:config(dba).
store_kai
2> kvs:version().
{version,"KVS KAI PURE XEN"}Create database for single node:
3> kvs:join().
[kvs] Mnesia Init
okYou can also create database by joining to existing cluster:
3> kvs:join('[email protected]').In that case you don't need to initialize the database to check table packages included into the schema:
4> kvs:dir().
[{table,"id_seq"},
{table,"subscription"},
{table,"feed"},
{table,"comment"},
{table,"entry"},
{table,"access"},
{table,"acl"},
{table,"user"}]Try to add some data:
1> rr(kvs_user).
2> kvs:put(#user{id="[email protected]"}).
ok
3> kvs:get(user,"[email protected]").
#user{id = "[email protected]",container = feed,...}
4> kvs:put(#user{id="[email protected]"}).
5> length(kvs:all(user)).
2The data in KVS represented as plain Erlang records. The first element of the tuple as usual indicates the name of bucket. And the second element usually corresponds to the index key field. Additional secondary indexes could be applied for stores that supports 2i, e.g. kai, mnesia, riak, mongodb.
All record could be chained into the double-linked lists in the database. So you can inherit from the ITERATOR record just like that:
-record(iterator, {id,version,
container,feed_id,prev,
next,feeds=[],guard,etc}).The layout of iterators are following:
> lists:zip(lists:seq(1,length((kvs:table(operation))#table.fields)),
(kvs:table(operation))#table.fields).
[{1,id},
{2,version},
{3,container},
{4,feed_id},
{5,prev},
{6,next},
{7,feeds},
{8,guard},
{9,etc},
{10,body},
{11,name},
{12,status}]This means your table will support add/remove linked list operations to lists.
1> kvs:add(#user{id="[email protected]"}).
2> kvs:add(#user{id="[email protected]"}).Read the chain (undefined means all)
3> kvs:entries(kvs:get(feed, user), user, undefined).
[#user{id="[email protected]"},
#user{id="[email protected]"}]Read flat values by all keys from table:
4> kvs:all(user).
[#user{id="[email protected]"},
#user{id="[email protected]"}]If you are using iterators records this automatically means you are using containers. Containers are just boxes for storing top/heads of the linked lists. Here is layout of containers:
-record(container, {id,top,count}).> lists:zip(lists:seq(1,length((kvs:table(feed))#table.fields)),
(kvs:table(feed))#table.fields).
[{1,id},
{2,top},
{3,count},
{4,aclver}]Usually you need only specify custom mnesia indexes and tables tuning. Riak, KAI and Redis backends don't need it. Group you table into table packages represented as modules with handle_notice API.
-module(kvs_feed).
-inclue_lib("kvs/include/metainfo.hrl").
metainfo() ->
#schema{name=kvs,tables=[
#table{name=feed,container=true,fields=record_info(fields,feed)},
#table{ name=entry,container=feed,fields=record_info(fields,entry),
keys=[feed_id,entry_id,from]},
#table{name=comment,container=feed,fields=record_info(fields,comment),
keys=[entry_id,author_id]} ]}.And plug it into schema config:
{kvs, {schema,[kvs_user,kvs_acl,kvs_feed,kvs_subscription]}},And on database init
1> kvs:join().It will create your custom schema.
Besides using KVS in production in a number of applications we have built on top of KVS several products. The first product is Chain Replication Database wit XA protocol. And second is social Feed Server for web shops and social sites.
The kvs semantic is totally compatible with XA protocol. Adding the object with PUT means only putting to database while ADD operations provides linking to the chain's container. Also linking operation LINK is provided separately.
dispatch({prepare,_,_,Tx}, #state{}) ->
kvs:info(?MODULE,"KVS PUT ~p:~p~n",[element(1,Tx),element(2,Tx)]),
kvs:put(Tx);
dispatch({commit,_,_,Tx}, #state{}) ->
kvs:info(?MODULE,"KVS LINK ~p:~p~n",[element(1,Tx),element(2,Tx)]),
kvs:link(Tx);
dispatch({rollback,_,_,Tx}, #state{}) ->
kvs:info(?MODULE,"KVS REMOVE ~p:~p~n",[element(1,Tx),element(2,Tx)]),
kvs:remove(Tx);See: https://github.com/spawnproc/cr
Here is Consumer behavior handlers of KVS FEEDS supervised processes
handle_notice( [kvs_feed,user,Owner,entry,Eid,add],
[#entry{feed_id=Fid}=Entry],
#state{feeds=Feeds}) ->
case lists:keyfind(Fid,2, S#state.feeds) of
false -> skip;
{_,_} -> add_entry(Eid,Fid,Entry) end,
{noreply, S};
handle_notice( [kvs_feed,user,Owner,entry,{Eid,FeedName},edit],
[#entry{feed_id=Fid}=Entry],
#state{feeds=Feeds}) ->
case lists:keyfind(FeedName,1,Feeds) of
false -> skip;
{_,Fid}-> update_entry(Eid,Fid,Entry) end,
{noreply, S};
handle_notice( [kvs_feed,user,Owner,entry,Eid,edit],
[#entry{feed_id=Fid}=Entry],
#state{feeds=Feeds}) ->
case lists:keyfind(Fid, 2, Feeds) of
false -> skip;
{_,_} -> update_entry(Eid,Fid,Entry) end,
{noreply, S};Here is the private implementation
add_entry(Eid,Fid,Entry) ->
E = Entry#entry{id = {Eid, Fid}, entry_id = Eid, feeds=[comments]},
Added = case kvs:add(E) of {error, Err} -> {error,Err}; {ok, En} -> En end,
msg:notify([kvs_feed, entry, {Eid, Fid}, added], [Added]).
update_entry(Eid,Fid,Entry) -> ...And that is how you can call it
kvs:notify([kvs_feed, user, "[email protected]", entry, Eid, add],
[#entry{}]).See: https://github.com/synrc/feeds
- Maxim Sokhatsky
- Andrii Zadorozhnii
- Vladimir Kirillov
- Alex Kalenuk
- Sergey Polkovnikov
- Andrey Martemyanov
OM A HUM