Getting Started:
Object Modeling with Erlang
To get started, let’s create the records that we’ll be using.
You can also download the code for this chapter at Github.
The Github version includes Erlang type specifications which have been omitted here for brevity.
%% msgy.hrl
-define(USER_BUCKET, <<"Users">>).
-define(MSG_BUCKET, <<"Msgs">>).
-define(TIMELINE_BUCKET, <<"Timelines">>).
-define(INBOX, "Inbox").
-define(SENT, "Sent").
-record(user, {user_name, full_name, email}).
-record(msg, {sender, recipient, created, text}).
-record(timeline, {owner, msg_type, msgs}).
We’ll be using the bucket Users
to store our data. We won’t be using bucket types here, so we don’t need to specify one.
To use these records to store data, we will first have to create a user
record. Then, when a user creates a message, we will append that message
to one or more timelines. If it’s a private message, we’ll append it to
the Recipient’s Inbox
timeline and to the User’s own Sent
timeline.
If it’s a group message, we’ll append it to the Group’s timeline, as
well as to the User’s Sent
timeline.
Buckets and keys revisited
Now that we’ve worked out how we will differentiate data in the system, let’s figure out our bucket and key names.
The bucket names are straightforward. We can use Users
, Msgs
, and
Timelines
. The key names, however, are a little more tricky. In past
examples we’ve used sequential integers, but this presents a problem: we
would need a secondary service to hand out these IDs. This service could
easily be a future bottleneck in the system, so let’s use a natural key.
Natural keys are a great fit for key/value systems because both humans
and computers can easily construct them when needed, and most of the
time they can be made unique enough for a KV store.
Bucket | Key Pattern | Example Key |
---|---|---|
Users |
<user_name> |
joeuser |
Msgs |
<username>_<datetime> |
joeuser_2014-03-06T02:05:13.223556Z |
Timelines |
<username>_<type>_<date> |
joeuser_Sent_2014-03-06Z marketing_group_Inbox_2014-03-06Z |
For the Users
bucket, we can be certain that we will want each
username to be unique, so let’s use the username
as the key. For the
Msgs
bucket, let’s use a combination of the username and the posting
datetime in an ISO 8601 Long
format. This combination gives us the pattern <username>_<datetime>
,
which produces keys like joeuser_2014-03-05T23:20:28Z
.
Now for Timelines
, we need to differentiate between Inbox
and Sent
timelines, so we can simply add that type into the key name. We will
also want to partition each collection object into some time period,
that way the object doesn’t grow too large (see note below).
For Timelines
, let’s use the pattern <username>_<type>_<date>
for
users, and <groupname>_Inbox_<date>
for groups, which will look like
joeuser_Sent_2014-03-06Z
or marketing_group_Inbox_2014-03-05Z
,
respectively.
Riak performs best with objects under 1-2 MB. Objects larger than that can hurt performance, especially if many siblings are being created. We will cover siblings, sibling resolution, and sibling explosions in the next chapter.
Keeping our story straight with repositories
Now that we’ve figured out our object model, let’s write some modules to act as repositories that will help us create and work with these records in Riak:
%% user_repository.erl
-module(user_repository).
-export([save_user/2,
get_user/2]).
-include("msgy.hrl").
save_user(ClientPid, User) ->
RUser = riakc_obj:new(?USER_BUCKET,
list_to_binary(User#user.user_name),
User),
riakc_pb_socket:put(ClientPid, RUser).
get_user(ClientPid, UserName) ->
{ok, RUser} = riakc_pb_socket:get(ClientPid,
?USER_BUCKET,
list_to_binary(UserName)),
binary_to_term(riakc_obj:get_value(RUser)).
%% msg_repository.erl
-module(msg_repository).
-export([create_msg/3,
get_msg/2]).
-include("msgy.hrl").
-spec create_msg(user_name(), user_name(), text()) -> msg().
create_msg(Sender, Recipient, Text) ->
#msg{sender=Sender,
recipient=Recipient,
created=get_current_iso_timestamp(),
text = Text}.
-spec get_msg(pid(), riakc_obj:key()) -> msg().
get_msg(ClientPid, MsgKey) ->
{ok, RMsg} = riakc_pb_socket:get(ClientPid,
?MSG_BUCKET,
MsgKey),
binary_to_term(riakc_obj:get_value(RMsg)).
%% @private
-spec get_current_iso_timestamp() -> datetimestamp().
get_current_iso_timestamp() ->
{_,_,MicroSec} = DateTime = erlang:now(),
{{Year,Month,Day},{Hour,Min,Sec}} = calendar:now_to_universal_time(DateTime),
lists:flatten(
io_lib:format("~4..0B-~2..0B-~2..0BT~2..0B:~2..0B:~2..0B.~6..0B",
[Year, Month, Day, Hour, Min, Sec, MicroSec])).
%% timeline_repository.erl
-module(timeline_repository).
-export([post_msg/2,
get_timeline/4]).
-include("msgy.hrl").
post_msg(ClientPid, Msg) ->
%% Save the canonical copy
SavedMsg = save_msg(ClientPid, Msg),
MsgKey = binary_to_list(riakc_obj:key(SavedMsg)),
%% Post to sender's Sent timeline
add_to_timeline(ClientPid, Msg, sent, MsgKey),
%% Post to recipient's Inbox timeline
add_to_timeline(ClientPid, Msg, inbox, MsgKey),
ok.
get_timeline(ClientPid, Owner, MsgType, Date) ->
TimelineKey = generate_key(Owner, MsgType, Date),
{ok, RTimeline} = riakc_pb_socket:get(ClientPid,
?TIMELINE_BUCKET,
list_to_binary(TimelineKey)),
binary_to_term(riakc_obj:get_value(RTimeline)).
%% --------------------------------------------------------------------
%% @private
save_msg(ClientPid, Msg) ->
MsgKey = Msg#msg.sender ++ "_" ++ Msg#msg.created,
ExistingMsg = riakc_pb_socket:get(ClientPid,
?MSG_BUCKET,
list_to_binary(MsgKey)),
SavedMsg = case ExistingMsg of
{error, notfound} ->
NewMsg = riakc_obj:new(?MSG_BUCKET, list_to_binary(MsgKey), Msg),
{ok, NewSaved} = riakc_pb_socket:put(ClientPid,
NewMsg,
[if_none_match, return_body]),
NewSaved;
{ok, Existing} -> Existing
end,
SavedMsg.
%% @private
add_to_timeline(ClientPid, Msg, MsgType, MsgKey) ->
TimelineKey = generate_key_from_msg(Msg, MsgType),
ExistingTimeline = riakc_pb_socket:get(ClientPid,
?TIMELINE_BUCKET,
list_to_binary(TimelineKey)),
UpdatedTimeline = case ExistingTimeline of
{error, notfound} ->
create_new_timeline(Msg, MsgType, MsgKey, TimelineKey);
{ok, Existing} ->
add_to_existing_timeline(Existing, MsgKey)
end,
{ok, SavedTimeline} = riakc_pb_socket:put(ClientPid,
UpdatedTimeline,
[return_body]),
SavedTimeline.
%% @private
create_new_timeline(Msg, MsgType, MsgKey, TimelineKey) ->
Owner = get_owner(Msg, MsgType),
Timeline = #timeline{owner=Owner,
msg_type=MsgType,
msgs=[MsgKey]},
riakc_obj:new(?TIMELINE_BUCKET, list_to_binary(TimelineKey), Timeline).
%% @private
add_to_existing_timeline(ExistingRiakObj, MsgKey) ->
ExistingTimeline = binary_to_term(riakc_obj:get_value(ExistingRiakObj)),
ExistingMsgList = ExistingTimeline#timeline.msgs,
UpdatedTimeline = ExistingTimeline#timeline{msgs=[MsgKey|ExistingMsgList]},
riakc_obj:update_value(ExistingRiakObj, UpdatedTimeline).
%% @private
get_owner(Msg, inbox) -> Msg#msg.recipient;
get_owner(Msg, sent) -> Msg#msg.sender.
%% @private
generate_key_from_msg(Msg, MsgType) ->
Owner = get_owner(Msg, MsgType),
generate_key(Owner, MsgType, Msg#msg.created).
%% @private
generate_key(Owner, MsgType, Date) when is_tuple(Date) ->
DateString = get_iso_datestamp_from_date(Date),
generate_key(Owner, MsgType, DateString);
generate_key(Owner, MsgType, Datetimestamp) ->
DateString = get_iso_datestamp_from_iso_timestamp(Datetimestamp),
MsgTypeString = case MsgType of
inbox -> ?INBOX;
sent -> ?SENT
end,
Owner ++ "_" ++ MsgTypeString ++ "_" ++ DateString.
%% @private
get_iso_datestamp_from_date(Date) ->
{Year,Month,Day} = Date,
lists:flatten(io_lib:format("~4..0B-~2..0B-~2..0B", [Year, Month, Day])).
%% @private
get_iso_datestamp_from_iso_timestamp(CreatedString) ->
{Date, _} = lists:split(10,CreatedString),
Date.
Finally, let’s test them:
%% msgy.erl
-module(msgy).
-export([main/0]).
-include("msgy.hrl").
main() ->
%% Setup our repositories
{ok, Pid} = riakc_pb_socket:start_link("127.0.0.1", 10017),
%% Create and save users
Joe = #user{user_name="joeuser",
full_name="Joe User",
email="joe.user@riak.com"},
Marleen = #user{user_name="marleenmgr",
full_name="Marleen Manager",
email="marleen.manager@riak.com"},
user_repository:save_user(Pid, Joe),
user_repository:save_user(Pid, Marleen),
%% Create new Msg, post to timelines
Msg = msg_repository:create_msg(Marleen#user.user_name, Joe#user.user_name, "Welcome to the company!"),
timeline_repository:post_msg(Pid, Msg),
%% Get Joe's inbox for today, get first message
{TodaysDate,_} = calendar:now_to_universal_time(erlang:now()),
JoesInboxToday = timeline_repository:get_timeline(Pid, Joe#user.user_name, inbox, TodaysDate),
JoesFirstMessage = msg_repository:get_msg(Pid, hd(JoesInboxToday#timeline.msgs)),
io:format("From: ~s~nMsg : ~s~n~n", [JoesFirstMessage#msg.sender, JoesFirstMessage#msg.text]),
ok.
As you can see, the repository pattern helps us with a few things:
- It helps us to see if an object exists before creating a new one
- It keeps our buckets and key names consistent
- It provides us with a consistent interface to work with.
While this set of repositories solves many of our problems, it is very minimal and doesn’t cover all the edge cases. For instance, what happens if two different people try to create a user with the same username?
We can also easily “compute” key names now, but how do we quickly look up the last 10 messages a user sent? Many of these answers will be application dependent. If your application shows the last 10 messages in reverse order, for example, you may want to store that set of data in another collection object to make lookup faster. There are drawbacks to every solution, but we recommend seeking out the key/value-based solution first, as it will likely be the quickest.
So to recap, in this chapter we learned:
- How to choose bucket names
- How to choose natural keys based on how we want to partition our data.