couchbase/couchbase-lite-java-ce-root

Replicator websocket connection timeout is not correctly handled on android

Closed this issue · 3 comments

workingenius commented

The code I read: 7481f1df5edfa1715fe8cb26b5930f38d48392e9 of couchbase-lite-java-common repo, here

  • A AbstractCBLWebSocket object is constructed with webSocket member var to be null, and it is not set until onOpen event listener is called when websocket handshake finishes (after the tcp connection establishes).
  • Couchbase lite core handles connection timeout (with a default value of 15), and it will eventually call AbstractCBLWebSocket.requestClose method to close it on timeout is detected.
  • Look at first several lines of this method, AbstractCBLWebSocket.requestClose, there's a shortcut return if webSocket is null, which is true when connection timeout happens. Okhttp know nothing about the request to close, it will keep trying (and retry) to connect.
  • And okhttp timeouts are explicitly set to 0 (because core would take charge), so it hangs for a very long time, maybe forever, even if the bad network recovers. Replicator is stuck in "connecting" state if we check from outside (sometimes "busy").
  • okhttp sometimes decides to stop and raise an exception in time, luckily, onFailure is called and end all this. (So we can know the failure from outside and restart replication)
  • There's another possibility that handshake finally succeeds after the core decide to close it because of timeout. I'm not sure what will happen then.

This is our analysis on the problem we encountered. Glad if you take a look.

bmeike commented

Hey! Thanks!

I've filed https://issues.couchbase.com/browse/CBL-1495 to track this.

bmeike commented

I believe this is fixed in couchbase-lite-java-common @ d2413f29730b1d9a4544244f