You shall not get excited

You Shall Not Get

Excited

Ivan Ribeiro Rochaivan.ribeiro@gmail.com

@irr

Agenda● Why not?

● ”History” about getting excited...

● What should I use...?

● My choice: Erlang● Learning from experience...

Why not?

”Once upon a time...”

”History” about getting excited...

(what is a ”technical” manager?)

…a ”technical”manager meets...

… a ”smart guy”

(a.k.a salesman)

Deal! You deserve it!

”no problem... we can fix it!”

”relax... the guys will change

it in the upcoming version...”

sometimes is not so easy to

fix all problems

”Ignorance more frequently begets

confidence than does

knowledge.”Charles Darwin

What should I use...?

Programming Languagesand tools...

● Languages

– Java, Ruby, Python, Perl, ...

– C/C++, Lisp, Haskell, Ocaml, Erlang, ...

● Web Servers

– Apache, nginx, ...

– Jetty/Tomcat, ...

– Yaws, inets, ...

Anyone using it?

Is it stable?

Does it scale?

(think about...)

You're not:

what means

scalable?- don’t design to scale infinitely

- consider 5X - 50X growth

- but > 100X requires redesign- break large systems into smaller services

Jeff Dean

http://bit.ly/clIJfL

no documentation...?● Good APIs?● Available books?● % Comments...?● Source code?

● Please! No ”magazines”...

still in doubt...?

Joe ArmstrongRobert Virding

Mike WilliamsClaes “Klacke” Wikström

and more...

My choice: Erlang

Learning from experience...

● > 3 years and learning...

● back end development

–REST API

– memcached API

– syslog API

– MySQL and Mnesia (data storage)

● functional + concurrent programming

● fault-tolerant and highly scalable

● very light-weight concurrency (processes)

● running massive systems for 20 years

● bit syntax and binaries (for protocol programming)

● links (encourages “let it crash” programming)

● ”concurrency belongs to the language and not the operating system”

Erlang(Why you should get excited...)

Concurrent Programming

– lots of processes

● the only way for processes to interact is

through message passing

–eventbased (epoll)

Erlang(Why you should get excited...)

No shared memory

– ”sharing is the property that preventsfault tolerance”

– destructive shared data modifications

do not occur!

Erlang(Why you should get excited...)

It's easy to test

and update your code

Erlang(Why you should get excited...)

HTTP/JSON support

– yaws (embedded)

– mochiweb (json)– misultin

–inets

– httpc

Erlang(Why you should get excited...)

many Libraries/tools for distributed programs...

– net_[adm|kernel] net_kernel:connect_node(N). net_adm:ping(N).

– rpc {R, E} = rpc:multicall(nodes(), Mod, Fun, [N], T), lists:foldl(fun(L, A) -> merge(A, [X || X <- L, not member(X, A)], N) end, C, R),

– epmd...

Erlang(Why you should get excited...)

”Mnesia is a distributed DataBase Management System

(DBMS), appropriate for telecommunications applications and other Erlang applications which require continuous operation

and exhibit soft real-time properties”

Mnesia is greatbut you should really know how to use it

and you MUST architecture your applicationto get the best results...

Erlang(Why you should get excited...)

OTP (Unix <-> C <==> OTP <-> Erlang)

”It’s an application operating system and a set of libraries and procedures used for building

large-scale, fault-tolerant, distributed applications”

– supervisor– applications

– gen_server– gen_fsm

– gen_event

”We should forget about small efficiencies, say about 97% of the time: premature optimization is the root of all evil.”

Donald Knuth

Erlang(caveats: You must be careful...)

Mnesia– can't handle very large data

–2 Gb

● ETS● DETS●MNESIA

Erlang(caveats: You must be careful...)

Mnesia–2 Gb fragments (mod (2^X))

– avoid rehash: create more fragments...● add● move● del*

Erlang(caveats: You must be careful...)

Mnesia–CAP theorem

–replica x partitioned networks

Erlang(caveats: You must be careful...)

Mnesia– need to check replica consistency?

● vector clocks... (avoid dependencies)

● should try timestamps

Erlang(caveats: You must be careful...)

Mnesia– event "running partitioned network"

● mnesia:set_master_nodes(L).

● must restart other nodes

– if some node will not recover soon...

● mnesia:force_load_table(T).

Erlang(caveats: You must be careful...)

Mnesia (QLC)● avoid retrieve large data sets● use cursors inside

transactions mnesia:activity(sync_transaction, fun(X) -> QC = qlc:cursor(X), QR = qlc:next_answers(QC, N), qlc:delete_cursor(QC), QR end, [qlc:q([E || E <- mnesia:table(Tab)])], mnesia_frag).

Erlang(caveats: You must be careful...)

Mnesia– load balance (read)

– avoid ”overload” one instance mnesia_lib:set({T, where_to_read}, Node)

Erlang(caveats: You must be careful...)

Logging– avoid overhead

● yaws● error_logger

– syslog● udp● tcp*● gen_server

+ handle_cast

Erlang(caveats: You must be careful...)

Messages– like a mailbox– per process (don't forget to read your messages)

Erlang(caveats: You must be careful...)

OTP–only API

– poor docs, books, articles...

Erlang(caveats: You must be careful...)

OTP– avoid ”synchronize” long calls

test(X) -> gen_server:call(?MODULE, {test, X}, 5000).

handle_call({test, X}, From, State) → spawn(fun() → %% do some long task gen_server:reply(From, Reply) end), {noreply, State}.

Erlang(caveats: You must be careful...)

Security– must be treated externally

● firewall

● private networks

– cookie based

-kernel inet_dist_use_interface {127,0,0,1}-kernel inet_dist_listen_min <min>-kernel inet_dist_listen_max <max>

> erl -kernel inet_dist_listen_min 9001 inet_dist_listen_max 9005(4369 TCP port, as it is used by epmd, ERL_EPMD_PORT environment variable)

Erlang(caveats: You must be careful...)

Code swapping/replacement– Be careful with spawn inside old (replaced) module... it will die!– The code of a module can exist in two variants in a system: current

and old (fully qualified function calls always refer to current code)

– If a third instance of the module is loaded, the code server will remove (purge) the old code and any processes

lingering in it will be terminated -module(m). -export([loop/0]).

loop() -> receive code_switch -> m:loop(); Msg -> % ...

loop()end.

Undocumented features...

you must be careful!

Erlang(caveats: You must be careful...)

Using async acceptors (prim_inet)– gen_tcp:accept(Listen)

Usages: https://github.com/irr/erl-tutorials/blob/master/ts/ts1/src/ts.erl

– prim_inet:async_accept(Socket, -1)Usages:https://github.com/irr/erl-tutorials/blob/master/ts/ts2/src/ts.erl

http://svn.apache.org/viewvc/incubator/thrift/trunk/thrift_server.erlhttp://www.trapexit.org/Building_a_Non-blocking_TCP_server_using_OTP_principles

”It is undocumented because it is an internal module that is not ment to be called from applications. Its interface may

change without warning in even the smallest patch.”

Erlang(caveats: You must be careful...)

Using socket in http modecase gen_tcp:listen(Port, [binary, {packet, http}, {reuseaddr, true}, {active, false}, {backlog, 30}]) of ...

case gen_tcp:recv(C#c.sock, 0, ?server_idle_timeout) of {ok, {http_header, _, 'Content-Length', _, Val}} ->... {error, {http_error, "\r\n"}} ->... {ok, http_eoh} ->...

Usage:http://www.trapexit.org/A_fast_web_server_demonstrating_some_undocumented_Erlang_features

Erlang(caveats: You must be careful...)

1> inet:ifget("eth0", [addr]).

{ok,[{addr,{192,168,1,101}}]}

2> inet:getiflist().

{ok,["lo","eth0"]}

3> inet_parse:ntoa({192,168,1,101}).

"192.168.1.101"

4> inet_parse:address("192.168.1.101").

{ok,{192,168,1,101}}

there is no

silverbullet!

Any

doubts?