This lib is intended to allow you to write an APNs provider for Apple Push Notificaion services (APNs) over HTTP2 in Erlang.
Copyright (c) 2017 Erlang Solutions Ltd. [email protected], released under the Apache 2 license
You can find the v1 here?
Note: Apns4erl v2
is still under development. Currently it supports push notifications with certificate and authentication token.
If you find any bugs or have a problem while using Apns4erl, please open an issue in this repo (or a pull request :)).
- You must have installed an updated Openssl version or, at least, be sure it supports TLS 1.2+. New APNs server only supports connections over TLS 1.2+.
- Erlang R19+
First we have to fill our config
data. There are two ways for do this, one is filling a config
file. This is an example you can find at test/test.config
:
{
apns,
[ {apple_host, "api.development.push.apple.com"}
, {apple_port, 443}
, {certfile, "priv/apns-dev-cert.pem"}
, {keyfile, "priv/apns-dev-key-noenc.pem"}
, {token_keyfile, "priv/APNsAuthKey_KEYID12345.p8"}
, {timeout, 10000}
%% APNs Headers
, {apns_id, undefined}
, {apns_expiration, 0}
, {apns_priority, 10}
, {apns_topic, "com.example.myapp"}
, {apns_collapse_id, undefined}
, {apns_push_type, "alert"}
%% Feedback
, {feedback_host, "feedback.push.apple.com"}
, {feedback_port, 2195}
]
]
}
The other way is send all that info as a parameter to apns:connect/1
function encapsulated in a apns_connection:connection()
structure:
#{ name := name()
, apple_host := host()
, apple_port := inet:port_number()
, certfile => path()
, keyfile => path()
, timeout => integer()
, type := type()
}.
APNs allows two connection types, one is using Provider Certificates
. The first certificate option is to supply cert paths in certfile
and keyfile
. Alternatively, you can supply a cert binary in certdata
and a keydata()
-type tuple (see: https://github.com/inaka/apns4erl/blob/master/src/apns_connection.erl#L64) in keydata
. Certs are the Provider Certificates
and the keys are the Private Key
both provided by Apple. We need them in .pem
format, here is an example of how to convert them, check the certificates section.
The other way to connect against APNs is using Provider Authentication Tokens
, for this choice you must fill the field token_keyfile
. This is a path to the Authentication Key provided by Apple. This is in .p8
format and it doesn't need conversion.
This key
will be needed in order to generate a token which will be used every time we try to push a notification. In connection's time it is not needed.
apns4erl
can be included as a dependency and started from yourapp.app.src
. You also can run it on the shell for testing.
> rebar3 compile
> erl -pa _build/default/lib/*/ebin -config test/test.config
Don't forget your config file if you want to use apns:connect/2
.
1> apns:start().
ok
After running apns4erl
app we can start creating connections. As we mentioned before there are two types of connections. Both are created using the functions apns:connect/1
and apns:connect/2
.
-
apns:connect/1
: This function accepts as a parameter anapns_connection:connection()
structure.#{ name := name() , apple_host := host() , apple_port := inet:port_number() , certdata => binary() , certfile => path() , keydata => keydata() , keyfile => path() , timeout => integer() , type := type() }.
where the
type
field indicates if iscertdata
,cert
, ortoken
. -
apns:connect/2
: The first argument is the type and the second one is the connection's name. In order to use it successfully we have to fill theconfig
file before, as explained inhow to use it?
section.
Example:
1> apns:connect(cert, my_first_connection).
{ok,<0.87.0>}
2> apns:connect(#{name => another_cert, apple_host => "api.push.apple.com", apple_port => 443,
certfile => "priv/cert.pem", keyfile => "priv/key.pem", type => cert}).
3> apns:connect(token, my_second_connection).
{ok,<0.95.0>}
Note cert
and token
define the type we want.
apns:connect/2
returns the connection pid
.
In some scenarios we don't want to assign names to the connections instead we want working just with the pid
(working with a pool of connections for example). If that is the case we use the same apns:connect/1
and apns:connect/2
functions but instead of a connection name we put undefined
:
1> apns:connect(cert, undefined).
{ok,<0.127.0>}
2> apns:connect(#{name => undefined, apple_host => "api.push.apple.com", apple_port => 443,
certfile => "priv/cert2.pem", keyfile => "priv/key2-noenc.pem", type => cert}).
{ok,<0.130.0>}
3> apns:connect(token, my_second_connection).
{ok,<0.132.0>}
In order to send Notifications over Provider Certificate
connection we will use apns:push_notification/3,4
.
We will need the connection, a notification, the device ID and some http2 headers. The connection is the atom
we used when we executed apns:connect/2
for setting a name or its pid
, the device ID is provided by Apple, the notification is a map
with the data we want to send, that map will be encoded to json later and the http2 headers can be explicitly sent as a parameter using apns:push_notification/4
or can be defined at the config
file, in that case we would use apns:push_notification/3
.
This is the headers
format:
-type headers() :: #{ apns_id => binary()
, apns_expiration => binary()
, apns_priority => binary()
, apns_topic => binary()
, apns_collapse_id => binary()
, apns_push_type => binary()
, apns_auth_token => binary()
}.
All of them are defined by Apple here
Lets send a Notification.
1> {ok, Pid} = apns:connect(cert, my_first_connection).
{ok,<0.85.0>}
2> DeviceId = <<"a0dc63fb059cb9c13b03e5c974af3dd33d67fed4147da8c5ada0626439e18935">>.
<<"a0dc63fb059cb9c13b03e5c974af3dd33d67fed4147da8c5ada0626439e18935">>
3> Notification = #{aps => #{alert => <<"you have a message">>}}.
#{aps => #{alert => <<"you have a message">>}}
4> apns:push_notification(my_first_connection, DeviceId, Notification).
{200,
[{<<"apns-id">>,<<"EFDE0D9D-F60C-30F4-3FF1-86F3B90BE434">>}],
no_body}
5> apns:push_notification(Pid, DeviceId, Notification).
{200,
[{<<"apns-id">>,<<"EFDE0D9D-F60C-30F4-3FF1-86F3B90BE654">>}],
no_body}
The result is the response itself, its format is:
-type response() :: { integer() % HTTP2 Code
, [term()] % Response Headers
, [term()] | no_body % Response Body
} | timeout.
And that's all.
This is the other way APNs allows us to send notifications. In this case we don't need a certificate but we will need a p8
file with the private key we will use to sign the token. Lets assume we've got the file APNsAuthKey_KEYID12345.p8
from Apple. We then have to fill the config
file key token_keyfile
with the path to that file.
We will need a kid
value, this is the key identifier. In our case is the last 10 chars of the file name (KEYID123456
). We will need also the iss
value, this is the Team Id, that can be checked on your Apple's Developer account, in our case it will be THEATEAM
. And that's it.
You can find more info here
In order to push a notification we will use apns:push_notification_token/4,5
. We will need the same attributes we used sending a notification over Provider Certificate
connections plus a signed token
. This token has a 1 hour life, so that means we can generate one token and use it many times till it expires. Lets try.
Create the token:
6> TeamId = <<"THEATEAM">>.
<<"THEATEAM">>
7> KeyID = <<"KEYID123456">>.
<<"KEYID123456">>
8> Token = apns:generate_token(TeamId, KeyID).
<<"eyJhbGciOiJFUzI1NiIsInR5cCI6IkpXVCIsImtpZCI6IktFWUlEMTIzNDU2In0.eyJpc3MiOiJUSEVBVEVBTSIsImlhdCI6MTQ4NjE0OTMzNH0.MEQC"...>>
Now push the notification:
12> DeviceId = <<"a0dc63fb059cb9c13b03e5c974af3dd33d67fed4147da8c5ada0626439e18935">>.
<<"a0dc63fb059cb9c13b03e5c974af3dd33d67fed4147da8c5ada0626439e18935">>
13> Notification = #{aps => #{alert => <<"you have a message">>}}.
#{aps => #{alert => <<"you have a message">>}}
14> apns:push_notification_token(my_second_connection, Token, DeviceId, Notification).
{200,
[{<<"apns-id">>,<<"EBC03BF9-A784-FDED-34F7-5A8D859DA977">>}],
no_body}
We can use this token for an entire hour, after that we will receive something like this:
16> apns:push_notification_token(my_second_connection, Token, DeviceId, Notification).
{403,
[{<<"apns-id">>,<<"03FF9497-8A6B-FFD6-B32B-160ACEDE35F0">>}],
[{<<"reason">>,<<"ExpiredProviderToken">>}]}
NOTE in order to push notifications, in both ways, we must call apns:push_notification/3,4
and apns:push_notification_token/4,5
from the same
process which created the connection. If we try to do it from a different one we will get an error {error, not_connection_owner}
.
If network goes down or something unexpected happens the gun
connection with APNs will go down. In that case apns4erl
will send a message {reconnecting, ServerPid}
to the client process, that means apns4erl
lost the connection and it is trying to reconnect. Once the connection has been recover a {connection_up, ServerPid}
message will be send.
We implemented an Exponential Backoff strategy. We can set the ceiling time adding the backoff_ceiling
variable on the config
file. By default it is set to 10 (seconds).
Apple recommends us to keep our connections open and avoid opening and closing very often. You can check the Best Practices for Managing Connections section.
But when closing a connection makes sense apns4erl
gives us the function apns:close_connection/1
where the parameter is the connection's name or the connection's pid
. After using it the name will be available for new connections again (if it was different than undefined
).
apns4erl
also allows us to get feedback from APNs service. It does it thru the binary API.
In order to get feedback we would need a Provider Certificate
. apns4erl
provides us two functions, apns:get_feedback/0
and apns:get_feedback/1
which require some Feedback's information like url, port, timeout... We can set that info in our config
file and use apns:get_feedback/0
. We can also send all that configuration as a parameter to apns:get_feedback/1
where the config structure must looks like this:
#{ host := string()
, port := pos_integer()
, certfile := string()
, keyfile => string()
, timeout := pos_integer()
}.
The response for both functions will be a list of feedback()
-type feedback() :: {calendar:datetime(), string()}.
Where the first element in the tuple is the date when the device uninstalled the app and the second element is the Device Id.
If you want to generate a new release of this project, you'll need to update the CHANGELOG.md file. We generally do it using github_changelog_generator
. This project needs a special option passed to it, tho: --exclude-tags-regex '1*'
. Otherwise, it will fail since version 2 releases started from a clean HEAD and therefore have nothing in common with the ones for version 1.