KVS: Erlang Abstract Term Database

Online Presentation: http://slid.es/maximsokhatsky/kvs

Features

Polymorphic Tuples
Managing Linked-Lists
Various Backends Support: Mnesia, Riak, KAI, Redis
Sequential Consistency via Feed Server
Basic Schema for Social Sites and Accounting
Extendable Schema
Supports Secondary Indexes for KAI, Mnesia and Riak
Change Backends on-the-fly
Supports Multiple backends at the same time
Xen Ready

Usage

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"} }}

Models

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

Applications

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.

Store Backends

Currently kvs includes following store backends:

Mnesia
Riak
KAI
Filesystem
Redis

Configuring

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
ok

You can also create database by joining to existing cluster:

3> kvs:join('kvs@synrc.com').

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"}]

Operations

Try to add some data:

1> rr(kvs).
2> kvs:put(#user{id="maxim@synrc.com"}).
ok
3> kvs:get(user,"maxim@synrc.com").
#user{id = "maxim@synrc.com",container = feed,...}
4> kvs:put(#user{id="doxtop@synrc.com"}).
5> length(kvs:all(user)).
2

Polymorphic Records

The 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.

Iterators

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="mes@ua.fm"}).
2> kvs:add(#user{id="dox@ua.fm"}).

Read the chain (undefined means all)

3> kvs:entries(kvs:get(feed, user), user, undefined).
[#user{id="mes@ua.fm"},
 #user{id="dox@ua.fm"}]

Read flat values by all keys from table:

4> kvs:all(user).
[#user{id="mes@ua.fm"},
 #user{id="dox@ua.fm"}]

Containers

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}]

Extending Schema

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/kvs.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.

Using KVS in real applications

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.

Chain Replication Database

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

Feeds Server

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, "maxim@synrc.com", entry, Eid, add],
           [#entry{}]).

See: https://github.com/synrc/feeds

Credits

Maxim Sokhatsky
Andrii Zadorozhnii
Vladimir Kirillov
Alex Kalenuk
Sergey Polkovnikov
Andrey Martemyanov

OM A HUM

cystbear / kvs