The client doesn't check for an active connection on its own, and instead relies on the network API response.
On a user command, the client prepares a REST request to the PubNub API and passes it to the networking API for processing. If there were no issues and the response arrived in time (the timeout is configurable), the results are parsed and wrapped into objects which are returned using callbacks.
If there was an issue during request processing, the networking API returns information about the error, which is then parsed and wrapped into objects and returned to the caller.
In case of network issues, realtime APIs return a status object with its category set to
There are also a set of other error categories:
PNNetworkIssuesCategory - no network connection when the client tried to send the request.
PNTLSConnectionFailedCategory - TLS handshake issues, generally due to poor connection quality and packet loss/delays.
PNTimeoutCategory - no response received before the specified timeout.
PNUnknownCategory - non-200 HTTP response code received from the server.
The Heartbeat API can be used as way to detect when network issues appear. Heartbeats send a request at a set interval (configured during client instance configuration) and use the observer's callback to report successful and failed attempts. If the network quality degrades between calls to this API, a socket with an active long-poll (subscribe API for realtime messages) may not notice those issues and end up listening for events on a broken pipe.
These status events may indicate network issues:
PNNetworkIssuesCategory. Swift has exclusive statuses that can reveal deeper information about the network from the browser APIs.
This status may arrive if a non-subscribe request doesn't receive a response from the server in time. This may happen only if:
- There are issues on the service itself
- There is a proxy between the client and the PubNub service, and the proxy doesn't send the request or return a response in time
- There are issues with poor network signal quality (or there is a lot of noise on the band)
No connection at the moment the request is sent. This status is returned immediately to the calling API through a callback.
- During subscription API usage, the client stumbled on a network issue or an unexpected response has been received.
- There were network issues and the networking API reported back to our SDK to generate error.
- service itself (may happen only if data provided to client has been malformed and break composed URI) or intermediate software (proxy server) returned response which can't be parsed by subscribe parser.
This status is exclusive to Swift and indicates that the browser reported network stability and the SDK was instructed to reconnect. This status is the first status fired after a connectivity disruption.
This status is exclusive to Swift and indicates that the browser reported network instability and the SDK has lost its connection to the host. This status is fired, but there are no guarantees on the chain of events (for example, the subscribe may time out first); in this situation, everything is dependent on the browser.
PNUnexpectedDisconnectCategory which can be generated in the middle of client lifecycle while it subscribed to real-time data channels.
In case that this status arrives, client starts pinging PubNub services using Time API every 10 seconds. As soon as a request is processed successfully, client will try to restore subscription on channels on which it was subscribed previously. If configured to keep timetoken on subscription restore, client will try to catch up on missed messages.
When client completes subscription restore after
PNUnexpectedDisconnectCategory it will generate
PNReconnectedCategory to notify observers what client has been reconnected to real-time data channels.
If more than 100 messages will arrive (from PubNub in-memory messages cache), it is better to use history API to fetch potentially missed messages. To be notified about this case,
requestMessageCountThreshold should be set to 100 - this will tell client to send
PubNub client allow user to describe behavior in case of network issue. For this purpose there is two parameters in
catchUpOnSubscriptionRestore. Both values by default set to
YES which configure client to restore subscription after it has been in unexpected disconnect state (because of network issues) and use last received time token to try catch up on missed messages.
restoreSubscription: instruct on whether client should restore subscription to channels after network issues or not. If set to
NOafter network failure client will clean up list of channels on which it subscribed and after network will go up again won't do anything.
catchUpOnSubscriptionRestore: instruct client whether after subscription restore (if
restoreSubscriptionis set to
YES) should use time token from previous subscribe loop (time token which has been received with last message/event while client has been connected). If property set to
YESclient will request and use new time token and all messages which has been sent while client was offline will be lost (sure they can be received with history calls).
keepTimeTokenOnListChange: this property not related to reconnection logic, but also may confuse users. If set to
YESclient will reuse previous time token when user subscribe to new channels. If set to
NOclient will use time token which will be received with subscription using new channels. If client has been subscribed to some channels with tt1 and subscribe to new channels, it will ignore tt2 (received in response for subscribe with new channels from PubNub service) and use tt1 to complete subscribe loop if set to
YES. This also mean, what if since tt1 in new channels was some messages - they will be received as well (this is what some users reported what
oldmessages has been received). If property set to
NOclient will use tt2 to complete subscribe loop.
The Apple ecosystem will begin pinging the time endpoint at regular intervals until the connection is restored or application goes into the background.
On some platforms, application owners have more visibility into networking conditions than the SDK. You can subscribe on same set of channels or empty list and client will terminate currently active long-poll request and start new one.
For the Apple SDKs, you can just subscribe to the same set of channels or an empty array to force a manual reconnect.
403 response on a subscribe event occurs when Access Manager is enabled and the user doesn't have access to the
channel or the
grant TTL has expired. In this case you need to create a new auth key, grant access to the user (from your server) and subscribe again.
For more info please check: http://www.pubnub.com/docs/cocoa-swift/pam-security#handling-permission-denied-errors
To completely disable PubNub's logger (even debug messages on client initialization) it is possible to set
Preprocessing Macros in Build Settings:
PUBNUB_DISABLE_LOGGER (just add keys itself w/o any values will work).
After macro added, code which is responsible for data output to stdout and file won't be compiled and calls to logger won't do anything.
If project has been added with CocoaPods, it is possible to add this macro automatically, if next code will be added into
post_install do |installer| installer.pods_project.targets.each do |target| target.build_configurations.each do |config| config.build_settings['GCC_PREPROCESSOR_DEFINITIONS'] ||= ['$(inherited)', 'PUBNUB_DISABLE_LOGGER'] if target.name =~ /^PubNub/ end end end
This hook will add macro for you, when dependency will be installed on
pod install or
pod update call.