Replicator Should Proactively Check Network Status

[bdkjones]

In the 3.2.4 SDK, when an Internet connection is lost, Replicator seems to use a back-off strategy to attempt to reconnect. The trouble with this approach is that once an Internet connection is restored, it's often a long while until the replicator is back online.

A Better Alternative:

Use https://github.com/ashleymills/Reachability.swift. The SDK can be immediately notified when Internet is available again and then reconnect the replicator instantly. It doesn't have to poll.

This also lets the SDK differentiate between "connection lost because the device has no Internet" and "connection lost because something sever-side has failed," which is a great advantage. In the former case (which is FAR more likely), polling is not needed and the replicator can simply wait until it's told that the device has Internet again. In the latter case, you can still use the back-off strategy to attempt reconnections.

This would be a great improvement/modernization for the SDK.

[bdkjones]

Another Alternative:

If incorporating modern network observation is too much work, you could also just expose a reconnect() API that lets the developer tell the Replicator: "try reconnecting RIGHT NOW." This isn't as nice, because then it's on the developer to handle network observations, but it would allow Couchbase's connectivity-handling to catch up to modern SDKs with a very minimal amount of engineering work on your end.

[velicuvlad]

This retry logic is shared across all CBL platforms. And so, using some external dependency that is not backed by Apple and is Swift specific won't do it. This is a long lasting topic that we've looked at from time to time, but there's some implications on the user end and scenarios we can't cover, if it were to offer a reconnect API. You could try tweak maxAttemptWaitTime depending on your use case.

[bdkjones]

You don’t need the dependency. Reachability is super simple and is just a nice wrapper around Apple’s existing APIs. Just lift the code and bring it into the SDK.

As for the “we don’t do anything that is platform-specific” reaction: why even have native SDKs then? This is the Swift SDK designed for Apple platforms. What does it matter if Android or Java don’t have a similar API for reconnecting instantly when network access is restored?

Why force the Swift SDK to wait 5 minutes when you can very easily detect the network availability change and reconnect instantly? Why can’t the Swift SDK just be better?

[pasin]

Actually CBL-iOS Replicator already has the reachability built in. When the replicator is offline due to a network or a transient error, the replicator will start its reachability monitor so that it could become online again when it can reach to the remote server immediately without needing to wait for the retry process to kick in (with exponential backoff).

The reachability was implemented several year ago by using SCNetworkReachability (I believe the new way is to use NetworkMonitor from the Network framework). If you are seeing some issues, it could be that there is a scenario that maybe SCNetworkReachability cannot detect that the remote server can be reached again or we have an issue that needs to be look at.

[bdkjones]

Hmm. I don’t see immediate reconnections when I toggle off WiFi, wait, and then turn it back on. The Swift SDK does not reconnect until the next exponential backoff threshold is reached, which can be minutes after the network is available again.

[pasin]

I have created CBL-7424 for investigation (CC: @velicuvlad).