Chapter 11. Getting Started with OTP

In order to help me understand how the gen_server behavior works, I drew the diagram shown in Figure 11-1.

Processing a call in gen_server
Figure 11-1. Processing a call in gen_server

The client does a gen_server:call(Server, Request). The server will then call the handle_call/3 function that you have provided in the Module that you told gen_server to use. gen_server will send your module the client’s Request, an identifier telling who the request is From, and the server’s current State.

Your handle_call/3 function will fulfill the client’s Request and send a {reply, Reply, NewState} tuple back to the server. It, in turn, will send the Reply back to the client, and use the NewState to update its state.

In Introducing Erlang and in the next two études, the client is you, using the shell. The module that handles the client’s call is contained within the same module as the gen_server framework, but, as the preceding diagram shows, it does not have to be.

You can learn more about working with OTP basics in Chapters 11 and 12 of Erlang Programming, Chapters 16 and 18 of Programming Erlang, Chapter 4 of Erlang and OTP in Action, and Chapters 14 through 20 of Learn You Some Erlang For Great Good!.

Étude 11-1: Get the Weather

In this étude, you will create a weather server using the gen_server OTP behavior.This server will handle requests using a four-letter weather station identifier and will return a brief summary of the weather. You may also ask the server for a list of most recently accessed weather stations.

Here is some sample output:

1> c(weather).
2> weather:start_link().
3> gen_server:call(weather, "KSJC").
{ok,[{location,"San Jose International Airport, CA"},
     {observation_time_rfc822,"Mon, 18 Feb 2013 13:53:00 -0800"},
     {temperature_string,"51.0 F (10.6 C)"}]}
4> gen_server:call(weather, "KITH").
{ok,[{location,"Ithaca / Tompkins County, NY"},
     {observation_time_rfc822,"Mon, 18 Feb 2013 16:56:00 -0500"},
     {weather,"A Few Clouds"},
     {temperature_string,"29.0 F (-1.6 C)"}]}
5> gen_server:call(weather,"NONE").
6> gen_server:cast(weather, "").
Most recent requests: ["KITH","KSJC"]

Obtaining Weather Data

To retrieve a web page, you must first call inets:start/0; you will want to do this in your init/1 code. Then, simply call httpc:request(url), where url is a string containing the URL you want. In this case, you will use the server provided by National Oceanic and Atmospheric Administration. This server accepts four-letter weather station codes and returns an XML file summarizing the current weather at that station. You request this data with a URL in the form

where NNNN is the station code.

If the call to httpc:request/1 fails you will get a tuple of the form {error,information}.

If it succeeds, you will get a tuple in the form:

{ok,{{"HTTP/1.1",code,"code message"},
     [{"HTTP header attribute","value"},
      {"Another attribute","another value"}],
     "page contents"}}

where code is the return code (200 means the page was found, 404 means it’s missing, anything else is some sort of error).

So, let’s say you have successfully retrieved a station’s data. You will then get page content that contains something like this:

<?xml version="1.0" encoding="ISO-8859-1"?>
<?xml-stylesheet href="latest_ob.xsl" type="text/xsl"?>
<current_observation version="1.0"
        <credit>NOAA's National Weather Service</credit>
                <title>NOAA's National Weather Service</title>
        <suggested_pickup>15 minutes after the hour</suggested_pickup>
        <location>San Jose International Airport, CA</location>
        <observation_time>Last Updated on Feb 18 2013, 11:53 am PST</observation_time>
  <observation_time_rfc822>Mon, 18 Feb 2013 11:53:00 -0800</observation_time_rfc822>
        <temperature_string>50.0 F (10.0 C)</temperature_string>
        <pressure_string>1017.7 mb</pressure_string>
        <dewpoint_string>43.0 F (6.1 C)</dewpoint_string>

Parsing the Data

You now have to parse that XML data. Luckily, Erlang comes with the xmerl_scan:string/1 function, which will parse your XML into a rather imposing-looking tuple. Here is what it looks like for a very simple bit of XML:

1> XML = "<pets><cat>Misha</cat><dog>Lady</dog></pets>".
3> Result = xmerl_scan:string(XML).

Ye cats! How you do work with that?! First, put this at the top of your code so that you can use xmerl's record definitions:


You can see all the details of the records at

Then, copy and paste this into your code. You could figure it out on your own, but that would take away from setting up the server, which is the whole point of this étude.

%% Take raw XML data and return a set of {key, value} tuples

analyze_info(WebData) ->
  %% list of fields that you want to extract
  ToFind = [location, observation_time_rfc822, weather, temperature_string],

  %% get just the parsed data from the XML parse result
  Parsed = element(1, xmerl_scan:string(WebData)),

  %% This is the list of all children under <current_observation>
  Children = Parsed#xmlElement.content,

  %% Find only XML elements and extract their names and their text content.
  %% You need the guard so that you don't process the newlines in the
  %% data (they are XML text descendants of the root element).
  ElementList = [{, extract_text(El#xmlElement.content)}
    || El <- Children, element(1, El) == xmlElement],

  %% ElementList is now a keymap; get the data you want from it.
  lists:map(fun(Item) -> lists:keyfind(Item, 1, ElementList) end, ToFind).

%% Given the parsed content of an XML element, return its first node value
%% (if it's a text node); otherwise return the empty string.

extract_text(Content) ->
  Item = hd(Content),
  case element(1, Item) of
    xmlText -> Item#xmlText.value;
    _ -> ""

Set up a Supervisor

Finally, you can easily crash the server by handing it a number instead of a string for the station code. Set up a supervisor to restart the server when it crashes.

1> c(weather_sup).
2> {ok, Pid} = weather_sup:start_link().
3> unlink(Pid).
4> gen_server:call(weather, "KGAI").
{ok,[{location,"Montgomery County Airpark, MD"},
     {observation_time_rfc822,"Mon, 18 Feb 2013 17:55:00 -0500"},
     {temperature_string,"37.0 F (3.0 C)"}]}
5> gen_server:call(weather, 1234).
** exception exit: {{badarg,[{erlang,'++',[1234,".xml"],[]},
     in function  gen_server:call/2 (gen_server.erl, line 180)

=INFO REPORT==== 18-Feb-2013::15:57:19 ===
    application: inets
    exited: stopped
    type: temporary
=ERROR REPORT==== 18-Feb-2013::15:57:19 ===
** Generic server weather terminating
** Last message in was 1234
** When Server state == ["KGAI"]
** Reason for termination ==
** {badarg,[{erlang,'++',[1234,".xml"],[]},

6> gen_server:call(weather, "KCMI").
{ok,[{location,"Champaign / Urbana, University of Illinois-Willard, IL"},
     {observation_time_rfc822,"Mon, 18 Feb 2013 16:53:00 -0600"},
     {weather,"Overcast and Breezy"},
     {temperature_string,"47.0 F (8.3 C)"}]}

See a suggested solution in Appendix A.

Étude 11-2: Wrapper Functions

In the previous étude, you made calls directly to gen_server. This is great for experimentation, but in a real application, you do not want other modules to have to know the exact format of the arguments you gave to gen_server:call/2 or gen_server:cast/2. Instead, you provide a "wrapper" function that makes the actual call. In this way, you can change the internal format of your server requests while the interface you present to other users remains unchanged.

In this étude, then, you will provide two wrapper functions report/1 and recent/0. The report/1 function will take a station name as its argument and do the appropriate gen_server:call; the recent/0 function will do an appropriate gen_server:cast. Everything else in your code will remain unchanged. You will, of course, have to add report/1 and recent/0 to the -export list.

Here’s some sample output.

1> c(weather).
2> weather:start_link().
3> weather:report("KSJC").
{ok,[{location,"San Jose International Airport, CA"},
     {observation_time_rfc822,"Tue, 26 Feb 2013 17:53:00 -0800"},
     {temperature_string,"56.0 F (13.3 C)"}]}
4> weather:report("XYXY").
5> weather:report("KCMI").
{ok,[{location,"Champaign / Urbana, University of Illinois-Willard, IL"},
     {observation_time_rfc822,"Tue, 26 Feb 2013 19:53:00 -0600"},
     {weather,"Light Rain Fog/Mist"},
     {temperature_string,"34.0 F (1.1 C)"}]}
6> weather:recent().
Most recent requests: ["KCMI","KSJC"]

See a suggested solution in Appendix A.

Étude 11-3: Independent Server and Client

In the previous études, the client and server have been running in the same shell. In this étude, you will make the server available to clients running in other shells.

To make a node available to other nodes, you need to name the node by using the -name option when starting erl. It looks like this:

michele@localhost $ erl -name serverNode
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)

This is a long name. You can also set up a node with a short name by using the -sname option:

michele@localhost $ erl -sname serverNode
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)

If you set up a node in this way, any other node can connect to it and do any shell commands at all. In order to prevent this, you may use the -setcookie Cookie when starting erl. Then, only nodes that have the same Cookie (which is an atom) can connect to your node.

To connect to a node, use the net_adm:ping/1 function, and give it the name of the server you want to connect to as its argument. If you connect succesfully, the function will return the atom pong; otherwise, it will return pang.

Here is an example. First, start a shell with a (very bad) secret cookie:

michele@localhost $ erl -sname serverNode -setcookie chocolateChip
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)

Now, open another terminal window, start a shell with a different cookie, and try to connect to the server node. I have purposely used a different user name to show that this works too.

steve@localhost $ erl -sname clientNode -setcookie oatmealRaisin
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)
(clientNode@localhost)1> net_adm:ping(serverNode@localhost).

The server node will detect this attempt and let you know about it:

=ERROR REPORT==== 28-Feb-2013::22:41:38 ===
** Connection attempt from disallowed node clientNode@localhost **

Quit the client shell, and restart it with a matching cookie, and all will be well.

steve@localhost erltest $ erl -sname clientNode -setcookie chocolateChip
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)
(clientNode@localhost)1> net_adm:ping(serverNode@localhost).

To make your weather report server available to other nodes, you need to do these things:

  • In the start_link/0 convenience method, set the first argument to gen_server:start_link/4 to {global, ?SERVER} instead of {local, ?SERVER}
  • In calls to gen_server:call/2 and gen_server:cast/2, replace the module name weather with {global, weather}
  • Add a connect/1 function that takes the server node name as its argument. This function will use net_adm:ping/1 to attempt to contact the server. It provides appropriate feedback when it succeeds or fails.

Here is what it looks like when one user starts the server in a shell.

michele@localhost $ erl -sname serverNode -setcookie meteorology
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)
(serverNode@localhost)1> weather:start_link().

And here’s another user in a different shell, calling upon the server.

steve@localhost $ erl -sname clientNode -setcookie meteorology
Erlang R15B02 (erts-5.9.2) [source] [smp:2:2] [async-threads:0] [hipe] [kernel-poll:false]

Eshell V5.9.2  (abort with ^G)
(clientNode@localhost)1> weather:connect(serverNode@localhost).
Connected to server.
(clientNode@localhost)2> weather:report("KSJC").
{ok,[{location,"San Jose International Airport, CA"},
     {observation_time_rfc822,"Thu, 28 Feb 2013 21:53:00 -0800"},
     {temperature_string,"52.0 F (11.1 C)"}]}
(clientNode@localhost)3> weather:report("KITH").
{ok,[{location,"Ithaca / Tompkins County, NY"},
     {observation_time_rfc822,"Fri, 01 Mar 2013 00:56:00 -0500"},
     {weather,"Light Snow"},
     {temperature_string,"31.0 F (-0.5 C)"}]}
(clientNode@localhost)4> weather:recent().

Whoa! What happened to the output from that last call? The problem is that the weather:recent/0 call does an io:format/3 call; that output will go to the server shell, since the server is running that code, not the client. Bonus points if you fix this problem by changing weather:recent/0 from using gen_server:cast/2 to use gen_server:call/2 instead to return the recently reported weather stations as its reply.

There’s one more question that went through my mind after I implemented my solution: how did I know that the client was calling the weather code running on the server and not the weather code in its own shell? It was easy to find out: I stopped the server.

User switch command
 --> q
michele@localhost $

Then I had the client try to get a weather report.

(clientNode@localhost)5> weather:report("KSJC").
** exception exit: {noproc,{gen_server,call,[{global,weather},"KSJC"]}}
     in function  gen_server:call/2 (gen_server.erl, line 180)

The fact that it failed told me that yes, indeed, the client was getting its information from the server.

See a suggested solution in Appendix A.

Étude 11-4: Chat Room

In the previous études, the client simply made a call to the server, and didn’t do any processing of its own. In this étude, you will create a "chat room" with a chat server and multiple clients, much as you see in Server with multiple clients.

Server with multiple clients. 

The interesting part of this program is that the client will also be a gen_server, as shown in Client as a gen_server.

Client as a gen_server

Up until now, you have been using a module name as the first argument to gen_server:call/2, and in the previous étude, you used net_adm:ping/1 to connect to a server.

In this étude, you won’t need net_adm:ping/1. Instead, you will use a tuple of the form {Module, Node} to directly connect to the node you want. So, for example, if you want to make a call to a module named chatroom on a node named lobby@localhost, you would do something like this:

gen_server:call({chatroom, }, Request)

This means you won’t need to connect with net_adm:ping/1.

Here is my design for the solution. You, of course, may come up with an entirely different and better design.

My solution has two modules, both of which use the gen_server behavior.

The chatroom Module

The first module, chatroom, will keep as its state a list of tuples, one tuple for each person in the chat. Each tuple has the format {{UserName, UserServer}, Pid}. The Pid is the one that gen_server:call receives in the From parameter; it’s guaranteed to be unique for each person in chat.

The handle_call/3 function will accept the following requests.

{login, UserName, ServerName}
Adds the user name, server name, and Pid (which is in the From parameter) to the server’s state. Don’t allow a duplicate user name from the same server.
Removes the user from the state list.
{say, Text}
Sends the given Text to all the other users in the chat room. Use gen_server:cast/2 to send the message.
Returns the list of names and servers for all people currently in the chat room.
{who, Person, ServerName}
Return the profile of the given person/server. (This is "extra credit"; see the following details about the person module). It works by calling the person module at ServerName and giving it a get_profile request.

The person Module

The other module, person, has a start_link/1 function; the argument is the node name of the chat room server. This will be passed on to the init/1 function. This is stored in the server’s state. I did this because many other calls need to know the chat room server’s name, and keeping it in the person’s state seemed a reasonable choice.

For extra credit, the state will also include the person’s profile, which is a list of {Key, Value} tuples.

The handle_call/3 takes care of these requests:

Returns the chat node name that’s stored in the server’s state. (Almost all of the wrapper functions to be described in the following section will need the chat node name.)
Returns the profile that’s stored in the server’s state (extra credit)
{set_profile, Key, Value}
If the profile already contains the key, replace it with the given value. Otherwise, add the key and value to the profile. Hint: use lists:keymember/3 and lists:keyreplace/4. (extra credit)

Because the chat room server uses gen_server:cast/2 to send messages to the people in the room, your handle_cast/3 function will receive messages sent from other users in this form:

{message, {FromUser, FromServer}, Text}

Wrapper Functions for the person module

A convenience function to get the name of the chat host node by doing gen_server:call(person, get_chat_node)
Calls the chat room server with a {login, UserName} request. If the user name is an atom, use atom_to_list/1 to convert it to a string.
Calls the chat room server with a logout request. As you saw in the description of chatroom, the server uses the process ID to figure out who should be logged out.
Calls the chat server with a {say, Text} request.
Calls the chat server with a users request.
who(UserName, UserNode)
Calls the chat server with a {who, UserName, UserNode} request to see the profile of the given person. (extra credit)
set_profile(Key, Value)
A convenience method that calls the person module with a {set_profile, Key, Value} request. (extra credit)

The login/2, logout/0, and say/2 wrapper functions do not call the chat server directly, because the from pid would be the process calling those functions (usually the shell), not the person server. Instead, these functions will make a gen_server:call to the person server. Its handle_call function will forward the gen_server:call to the chatroom. That way, the chat room server sees the request coming from the person server.

Putting it All Together

Here is what the chat room server looks like. The lines beginning with Recipient list: are debug output. I have gotten rid of the startup lines from the erl command.

erl -sname lobby

(lobby@localhost)1> chatroom:start_link().
Recipient list: [{"Steve",sales@localhost},{"Michele",marketing@localhost}]
Recipient list: [{"David",engineering@localhost},
Recipient list: [{"David",engineering@localhost},{"Steve",sales@localhost}]
Recipient list: [{"David",engineering@localhost},

And here are three other servers talking to one another and setting profile information.

erl -sname sales

(sales@localhost)1> person:start_link(lobby@localhost).
Chat node is: lobby@localhost
(sales@localhost)2> person:login("Steve").
{ok,"Logged in."}
(sales@localhost)3> person:set_profile(city, "Chicago").
David (engineering@localhost) says: "Hi, everyone."
(sales@localhost)4> person:say("How's things in Toronto, David?").
Michele (marketing@localhost) says: "New product launch is next week."
(sales@localhost)5> person:say("oops, gotta run.").
(sales@localhost)6> person:logout().
erl -sname engineering

(engineering@localhost)1> person:start_link(lobby@localhost).
Chat node is: lobby@localhost
(engineering@localhost)2> person:login("David").
{ok,"Logged in."}
(engineering@localhost)3> person:set_profile(city, "Toronto").
(engineering@localhost)4> person:set_profile(department, "New Products").
{ok,[{department,"New Products"},{city,"Toronto"}]}
(engineering@localhost)5> person:say("Hi, everyone.").
Steve (sales@localhost) says: "How's things in Toronto, David?"
Michele (marketing@localhost) says: "New product launch is next week."
(engineering@localhost)6> person:users().
Steve (sales@localhost) says: "oops, gotta run."
erl -sname marketing

(marketing@localhost)1> person:start_link(lobby@localhost).
Chat node is: lobby@localhost
(marketing@localhost)2> person:login("Michele").
{ok,"Logged in."}
(marketing@localhost)3> person:set_profile(city, "San Jose").
{ok,[{city,"San Jose"}]}
David (engineering@localhost) says: "Hi, everyone."
Steve (sales@localhost) says: "How's things in Toronto, David?"
(marketing@localhost)4> person:say("New product launch is next week.").
Steve (sales@localhost) says: "oops, gotta run."
(marketing@localhost)5> person:users().

See a suggested solution in Appendix A.