Home | History | Annotate | Download | only in gcm
      1 page.title=GCM Advanced Topics
      2 @jd:body
      3 
      4 <div id="qv-wrapper">
      5 <div id="qv">
      6 
      7 <h2>Quickview</h2>
      8 
      9 <ul>
     10 <li>Learn more about GCM advanced features.</li>
     11 </ul>
     12 
     13 
     14 <h2>In this document</h2>
     15 
     16 <ol>
     17 <li><a href="#lifetime">Lifetime of a Message</a></li>
     18 <li><a href="#throttling">Throttling</a></li>
     19 <li><a href="#reg-state">Keeping the Registration State in Sync</a>
     20   <ol>
     21     <li><a href="#canonical">Canonical IDs</a></li>
     22   </ol>
     23 </li>
     24 <li><a href="#retry">Automatic Retry Using Exponential Back-Off</a></li>
     25 <li><a href="#unreg">Unregistration</a>
     26   <ol>
     27     <li><a href="#unreg-why">Why you should rarely unregister</a></li>
     28     <li><a href="#unreg-how">How unregistration works</a></li>
     29   </ol>
     30 </li>
     31 <li><a href="#collapsible">Send-to-Sync vs. Messages with Payload</a>
     32   <ol>
     33     <li><a href="#s2s">Send-to-sync messages</a></li>
     34     <li><a href="#payload">Messages with payload</a></li>
     35 <li><a href="#which">Which should I use?</a></li>
     36     </ol>
     37 </li>
     38 <li><a href="#ttl">Setting an Expiration Date for a Message</a> </li>
     39 <li><a href="#throttling"></a><a href="#multi-senders">Receiving Messages from
     40 Multiple Senders</a></li>
     41 </ol>
     42 
     43 </div>
     44 </div>
     45 <p>This document covers advanced topics for GCM.</p>
     46 
     47 
     48 
     49 
     50 <h2 id="msg-lifetime">Lifetime of a Message</h2>
     51 <p>When a 3rd-party server posts a message to GCM and receives a message ID back,
     52 it does not mean that the message was already delivered to the device. Rather, it
     53 means that it was accepted for delivery. What happens to the message after it is
     54 accepted depends on many factors.</p>
     55 
     56 <p>In the best-case scenario, if the device is connected to GCM, the screen is on,
     57 and there are no throttling restrictions (see <a href="#throttling">Throttling</a>),
     58 the message will be delivered right away.</p>
     59 
     60 <p>If the device is connected but idle, the message will still be
     61 delivered right away unless the <code>delay_while_idle</code> flag is set to true.
     62 Otherwise, it will be stored in the GCM servers until the device is awake. And
     63 that's where the <code>collapse_key</code> flag plays a role: if there is already
     64 a message with the same collapse key (and registration ID) stored and waiting for
     65 delivery, the old message will be discarded and the new message will take its place
     66 (that is, the old message will be collapsed by the new one). However, if the collapse
     67 key is not set, both the new and old messages are stored for future delivery.
     68 Collapsible messages are also called <a href="#s2s">send-to-sync messages</a>.</p>
     69 
     70 <p class="note"><strong>Note:</strong> There is a limit on how many messages can
     71 be stored without collapsing. That limit is currently 100. If the limit is reached,
     72 all stored messages are discarded. Then when the device is back online, it receives
     73 a special message indicating that the limit was reached. The application can then
     74 handle the situation properly, typically by requesting a full sync.
     75 <br><br>
     76 Likewise, there is a limit on how many <code>collapse_key</code>s you can have for
     77 a particular device. GCM allows a maximum of 4 different collapse keys to be used
     78 by the GCM server per device
     79 any given time. In other words, the GCM server can simultaneously store 4 different
     80 send-to-sync messages, each with a different collapse key. If you exceed this number
     81 GCM will only keep 4 collapse keys, with no guarantees about which ones they will be.
     82 See <a href="#s2s">Send-to-sync messages</a> for more information.
     83 </p>
     84 
     85 <p>If the device is not connected to GCM, the message will be stored until a
     86 connection is established (again respecting the collapse key rules). When a connection
     87 is established, GCM will deliver all pending messages to the device, regardless of
     88 the <code>delay_while_idle</code> flag. If the device never gets connected again
     89 (for instance, if it was factory reset), the message will eventually time out and
     90 be discarded from GCM storage. The default timeout is 4 weeks, unless the
     91 <code>time_to_live</code> flag is set.</p>
     92 
     93 <p>Finally, when GCM attempts to deliver a message to the device and the
     94 application was uninstalled, GCM will discard that message right away and
     95 invalidate the registration ID. Future attempts to send a message to that device
     96 will get a <code>NotRegistered</code> error. See <a href="#unreg">
     97 How Unregistration Works</a> for more information.</p>
     98 <p>Although is not possible to track the status of each individual message, the
     99 Google Cloud Console stats are broken down by messages sent to device, messages
    100 collapsed, and messages waiting for delivery.</p>
    101 
    102 <h2 id="throttling">Throttling</h2>
    103 <p>To prevent abuse (such as sending a flood of messages to a device) and
    104 to optimize for the overall network efficiency and battery life of
    105 devices, GCM implements throttling of messages using a token bucket
    106 scheme. Messages are throttled on a per application and per <a href="#collapsible">collapse
    107 key</a> basis (including non-collapsible messages). Each application
    108 collapse key is granted some initial tokens, and new tokens are granted
    109 periodically therefter. Each token is valid for a single message sent to
    110 the device. If an application collapse key exhausts its supply of
    111 available tokens, new messages are buffered in a pending queue until
    112 new tokens become available at the time of the periodic grant. Thus
    113 throttling in between periodic grant intervals may add to the latency
    114 of message delivery for an application collapse key that sends a large
    115 number of messages within a short period of time. Messages in the pending
    116 queue of an application collapse key may be delivered before the time
    117 of the next periodic grant, if they are piggybacked with messages
    118 belonging to a non-throttled category by GCM for network and battery
    119 efficiency reasons.</p>
    120 
    121 <h2 id="reg-state">Keeping the Registration State in Sync</h2>
    122 <p>Whenever the application registers as described in
    123 <a href="{@docRoot}google/gcm/client.html">Implementing GCM Client</a>,
    124 it should save the registration ID for future use, pass it to the
    125 3rd-party server to complete the registration, and keep track of
    126 whether the server completed the registration. If the server fails
    127 to complete the registration, it should try again or unregister from GCM.</p>
    128 
    129 <p>There are also two other scenarios that require special care:</p>
    130 <ul>
    131   <li>Application update</li>
    132   <li>Backup and restore
    133   </li>
    134 </ul>
    135 <p>When an application is updated, it should invalidate its existing registration
    136 ID, as it is not guaranteed to work with the new version.  Because there is no
    137 lifecycle method called when the application is updated, the best way to achieve
    138 this validation is by storing the current application version when a registration
    139 ID is stored. Then when the application is started, compare the stored value with
    140 the current application version. If they do not match, invalidate the stored data
    141 and start the registration process again.</p>
    142 
    143 <p>Similarly, you should not save the registration ID when an application is
    144 backed up. This is because the registration ID could become invalid by the time
    145 the application is restored, which would put the application in an invalid state
    146 (that is, the application thinks it is registered, but the server and GCM do not
    147 store that registration ID anymore&mdash;thus the application will not get more
    148 messages).</p>
    149 <h3 id="canonical">Canonical IDs</h3>
    150 <p>On the server side, as long as the application is behaving well, everything
    151 should work normally. However, if a bug in the application triggers multiple
    152 registrations for the same device, it can be hard to reconcile state and you might
    153 end up with duplicate messages.</p>
    154 <p>GCM provides a facility called &quot;canonical registration IDs&quot; to easily
    155 recover from these situations. A canonical registration ID is defined to be the ID
    156 of the last registration requested by your application. This is the ID that the
    157 server should use when sending messages to the device.</p>
    158 <p>If later on you try to send a message using a different registration ID, GCM
    159 will process the request as usual, but it will include the canonical registration
    160 ID in the <code>registration_id</code> field of the response. Make sure to replace
    161 the registration ID stored in your server with this canonical ID, as eventually
    162 the ID you're using will stop working.</p>
    163 
    164 <h2 id="retry">Automatic Retry Using Exponential Back-Off</h2>
    165 
    166 <p>When registration or unregistration fails, the app should retry the failed operation.</p>
    167 <p>In the simplest case, if your application attempts to register and GCM is not a
    168 fundamental part of the application, the application could simply ignore the error
    169 and try to register again the next time it starts. Otherwise, it should retry the
    170 previous operation using exponential back-off. In exponential back-off, each time
    171 there is a failure, it should wait twice the previous amount of time before trying
    172 again. If the register (or unregister) operation was synchronous, it could be retried
    173 in a simple loop. However, since it is asynchronous, the best approach is to schedule
    174 a {@link android.app.PendingIntent} to retry the operation.
    175 
    176 <h2 id="unreg">Unregistration</h2>
    177 
    178 <p>This section explains when you should unregister in GCM and what happens
    179 when you do.</p>
    180 
    181 <h3 id="unreg-why">Why you should rarely unregister</h3>
    182 
    183 <p>A registration ID (regID) represents a particular Android application running
    184 on a particular device. You should only need to unregister in rare cases, such as
    185 if you want an app to stop receiving messages, or if you suspect that the regID has
    186 been compromised. In general, though, once an app has a regID, you shouldn't need
    187 to change it.</p>
    188 
    189 <p>In particular, you should never unregister your app as a mechanism for
    190 logout or for switching between users, for the following reasons:</p>
    191 
    192 <ul>
    193   <li>A regID maps an app to a device. It isn't associated with a particular
    194   logged in user. If you unregister and then re-register, GCM may return the same
    195   ID or a different ID&mdash;there's no guarantee either way.</li>
    196 
    197   <li>Unregistration may take up to 5 minutes to propagate.</li>
    198   <li>After unregistration, re-registration may again take up to 5 minutes to
    199 propagate. During this time messages may be rejected due to the state of being
    200 unregistered, and after all this, messages may still go to the wrong user.</li>
    201 </ul>
    202 
    203 
    204 <p>The solution is to manage your own mapping between users, the regID, and
    205 individual messages:</p>
    206 
    207 <ul>
    208   <li>Your app server should maintain a mapping between the current user
    209 and the regID. This should include information about which user is supposed to
    210 receive a particular message.</li>
    211   <li>The app running on the device should check to ensure that messages it
    212 receives match the logged in user.</li>
    213 </ul>
    214 
    215 
    216 <h3 id="unreg-how">How unregistration works</h3>
    217 
    218 <p>An application can be automatically unregistered after it is uninstalled from
    219 the device. However, this process does not happens right away, as Android does not
    220 provide an uninstall callback. What happens in this scenario is as follows:</p>
    221 <ol>
    222   <li>The end user uninstalls the application.</li>
    223   <li>The 3rd-party server sends a message to GCM server.</li>
    224   <li>The GCM server sends the message to the device.</li>
    225   <li>The GCM client receives the message and queries Package Manager about
    226 whether there are broadcast receivers configured to receive it, which returns
    227 <code>false</code>.
    228 </li>
    229   <li>The GCM client informs the GCM server that the application was uninstalled.</li>
    230   <li>The GCM server marks the registration ID for deletion.</li>
    231   <li>The 3rd-party server sends a message to  GCM.</li>
    232   <li>The GCM returns a <code>NotRegistered</code> error message to the 3rd-party server.</li>
    233   <li>The 3rd-party deletes the registration ID.
    234   </li>
    235 </ol>
    236 
    237 <p class ="note"><strong>Note:</strong> The GCM client is the Google Cloud
    238 Messaging framework present on the device.</p>
    239 
    240 <p>Note that it might take a while for the registration ID be completely removed
    241 from GCM. Thus it is possible that messages sent during step 7 above gets a valid
    242 message ID as response, even though the message will not be delivered to the device.
    243 Eventually, the registration ID will be removed and the server will get a
    244 <code>NotRegistered</code> error, without any further action being required from
    245 the 3rd-party server (this scenario happens frequently while an application is
    246 being developed and tested).</p>
    247 
    248 <h2 id="collapsible">Send-to-Sync  vs. Messages with Payload</h2>
    249 
    250 <p>Every message sent in GCM has the following characteristics:</p>
    251 <ul>
    252   <li>It has a payload limit of 4096 bytes.</li>
    253   <li>By default, it is stored by GCM for 4 weeks.</li>
    254 </ul>
    255 
    256 <p>But despite these similarities, messages can behave very differently depending
    257 on their particular settings. One major distinction between messages is whether
    258 they are collapsed (where each new message replaces the preceding message) or not
    259 collapsed (where each individual message is delivered). Every message sent in GCM
    260 is either a &quot;send-to-sync&quot; (collapsible) message or a &quot;message with
    261 payload&quot; (non-collapsible message). These concepts are described in more
    262 detail in the following sections.</p>
    263 
    264 <h3 id="s2s"><strong>Send-to-sync messages</strong></h3>
    265 
    266 <p>A send-to-sync (collapsible) message is often a &quot;tickle&quot; that tells
    267 a mobile application to sync data from the server. For example, suppose you have
    268 an email application. When a user receives new email on the server, the server
    269 pings the mobile application with a &quot;New mail&quot; message. This tells the
    270 application to sync to the server to pick up the new email. The server might send
    271 this message multiple times as new mail continues to accumulate, before the application
    272 has had a chance to sync. But if the user has received 25 new emails, there's no
    273 need to preserve every &quot;New mail&quot; message. One is sufficient. Another
    274 example would be a sports application that updates users with the latest score.
    275 Only the most recent message is relevant, so it makes sense to have each new
    276 message replace the preceding message. </p>
    277 
    278 <p>The email and sports applications are cases where you would probably use the
    279 GCM <code>collapse_key</code> parameter. A <em>collapse key</em> is an arbitrary
    280 string that is used to collapse a group of like messages when the device is offline,
    281 so that only the most recent message gets sent to the client. For example,
    282 &quot;New mail,&quot; &quot;Updates available,&quot; and so on</p>
    283 <p>GCM allows a maximum of 4 different collapse keys to be used by the GCM server
    284 at any given time. In other words, the GCM server can simultaneously store 4
    285 different send-to-sync messages per device, each with a different collapse key.
    286 For example, Device A can have A1, A2, A3, and A4. Device B can have B1, B2, B3,
    287 and B4, and so on. If you exceed this number GCM will only keep 4 collapse keys, with no
    288 guarantees about which ones they will be.</p>
    289 
    290 <h3 id="payload">Messages with payload</h3>
    291 <p>Unlike a send-to-sync message, every &quot;message with payload&quot;
    292 (non-collapsible message) is delivered. The payload the message contains can be
    293 up to 4kb. For example, here is a JSON-formatted message in an IM application in
    294 which spectators are discussing a sporting event:</p>
    295 
    296 <pre class="prettyprint pretty-json">{
    297   "registration_id" : "APA91bHun4MxP5egoKMwt2KZFBaFUH-1RYqx...",
    298   "data" : {
    299     "Nick" : "Mario",
    300     "Text" : "great match!",
    301     "Room" : "PortugalVSDenmark",
    302   },
    303 }</pre>
    304 
    305 <p>A &quot;message with payload&quot; is not simply a &quot;ping&quot; to the
    306 mobile application to contact the server to fetch data. In the aforementioned IM
    307 application, for example, you would want to deliver every message, because every
    308 message has different content. To specify a non-collapsible message, you simply
    309 omit the <code>collapse_key</code> parameter. Thus GCM will send each message
    310 individually. Note that the order of delivery is not guaranteed.</p>
    311 
    312 <p>GCM will store up to 100 non-collapsible messages. After that, all messages
    313 are discarded from GCM, and a new message is created that tells the client how
    314 far behind it is. The message is delivered through a regular
    315 <code>com.google.android.c2dm.intent.RECEIVE</code> intent with the
    316 extra <code>message_type</code>, for which the value is always the string
    317 &quot;deleted_messages&quot;.</p>
    318 
    319 <p>The application should respond by syncing with the server to recover the
    320 discarded messages. </p>
    321 
    322 <h3 id="which">Which should I use?</h3>
    323   <p>If your application does not need to use non-collapsible messages, collapsible
    324 messages are a better choice from a performance standpoint, because they put less
    325 of a burden on the device battery. However, if you use collapsible messages, remember that
    326 <strong>GCM only allows a maximum of 4 different collapse keys to be used by the GCM server
    327 per device at any given time</strong>. You must not exceed this number, or it could cause
    328 unpredictable consequences.</p>
    329 
    330 <h2 dir="ltr" id="ttl">Setting an Expiration Date for a Message</h2>
    331 <p>The Time to Live (TTL) feature lets  the sender  specify the maximum lifespan
    332 of a message using the <code>time_to_live</code> parameter in the send request.
    333 The value of this parameter must be a duration from 0 to 2,419,200 seconds, and
    334 it corresponds to the maximum period of time for which GCM will store and try to
    335 deliver the message. Requests that don't contain this field default to the maximum
    336 period of 4 weeks.</p>
    337 <p>Here are some possible uses for this feature:</p>
    338 <ul>
    339   <li>Video chat incoming calls</li>
    340   <li>Expiring invitation events</li>
    341   <li>Calendar events</li>
    342 </ul>
    343 <h3 id="bg">Background </h3>
    344 <p>GCM will usually deliver messages immediately after they are sent. However,
    345 this might not always be possible. For example, the device could be turned off,
    346 offline, or otherwise unavailable. In other cases, the sender itself might request
    347 that messages not be delivered until the device becomes active by using the
    348 <code>delay_while_idle</code> flag. Finally, GCM might intentionally delay messages
    349 to prevent an application from consuming excessive resources and negatively
    350 impacting battery life.</p>
    351 
    352 <p>When this happens, GCM will store the message and deliver it as soon as it's
    353 feasible. While this is fine in most cases, there are some applications for which
    354 a late message might as well never be delivered. For example, if the message is
    355 an incoming call or video chat notification, it will only be meaningful for a
    356 small period of time before the call is terminated. Or if the message is an
    357 invitation to an event, it will be useless if received after the event has ended.</p>
    358 
    359 <p>Another advantage of specifying the expiration date for a message is that GCM
    360 will never throttle messages with a <code>time_to_live</code> value of 0 seconds.
    361 In other words, GCM will guarantee best effort for messages that must be delivered
    362 &quot;now or never.&quot; Keep in mind that a <code>time_to_live</code> value of
    363 0 means messages that can't be delivered immediately will be discarded. However,
    364 because such messages are never stored, this provides the best latency for
    365 sending notifications.</p>
    366 
    367 <p>Here is an example of a JSON-formatted request that includes TTL:</p>
    368 <pre class="prettyprint pretty-json">
    369 {
    370   "collapse_key" : "demo",
    371   "delay_while_idle" : true,
    372   "registration_ids" : ["xyz"],
    373   "data" : {
    374     "key1" : "value1",
    375     "key2" : "value2",
    376   },
    377   "time_to_live" : 3
    378 },
    379 </pre>
    380 
    381 
    382 <h2 id="multi-senders">Receiving Messages from Multiple Senders</h2>
    383 
    384 <p>GCM allows multiple parties to send messages to the same application. For
    385 example, suppose your application is an articles aggregator with multiple
    386 contributors, and you want each of them to be able to send a message when they
    387 publish a new article. This message might contain a URL so that the application
    388 can download the article. Instead of having to centralize all sending activity in
    389 one location, GCM gives you the ability to let each of these contributors send
    390 its own messages.</p>
    391 
    392 <p>To make this possible, all you need to do is have each sender generate its own
    393 project number. Then include those IDs in the sender field, separated by commas,
    394 when requesting a registration. Finally, share the registration ID with your
    395 partners, and they'll be able to send messages to your application using their
    396 own authentication keys.</p>
    397 <p>This code snippet illustrates this feature. Senders are passed as an intent
    398 extra in a comma-separated list:</p>
    399 
    400 <pre class="prettyprint pretty-java">Intent intent = new Intent(GCMConstants.INTENT_TO_GCM_REGISTRATION);
    401 intent.setPackage(GSF_PACKAGE);
    402 intent.putExtra(GCMConstants.EXTRA_APPLICATION_PENDING_INTENT,
    403         PendingIntent.getBroadcast(context, 0, new Intent(), 0));
    404 String senderIds = &quot;968350041068,652183961211&quot;;
    405 intent.putExtra(GCMConstants.EXTRA_SENDER, senderIds);
    406 ontext.startService(intent);
    407  </pre>
    408 
    409 <p>Note that there is limit of 100 multiple senders.</p>
    410