Socketnetwork timeout and polling #411
Conversation
| @@ -137,7 +137,7 @@ private Map<Integer, Socket> connectClient(final NetworkConfiguration conf) | |||
| } catch (ConnectException e) { | |||
| // A ConnectionException is expected if the opposing side is not listening for our | |||
| // connection attempt yet. We ignore this and try again. | |||
| Thread.sleep(1L << ++attempts); | |||
| Thread.sleep(Math.min(1 << attempts++, 5_000)); | |||
There was a problem hiding this comment.
The choice of an upper bound is based on the assumption that we always want to attempt to connect frequently - and 5 seconds was picked based on our own observations.
Alternatively this upper bound could be made configurable, or the entire calculation could be configurable - e.g. some Function<Integer, Long> pollInterval with the default being attempts -> 1 << attemps
There was a problem hiding this comment.
It should probably just keep waiting the same amount, scale linearly, or max out after a period. We probably never want to wait 24 days (2^31 ms) between retries. But the existing implementation is a bit weird. 5 seconds is probably a good amount.
|
I assume the build failure is due to missing permissions on the job https://github.com/aicis/fresco/blob/master/.github/workflows/test-report.yml ? (Similar to 80c3c9b ? ) |
|
Yea, GitHub actions have a tendency to break with forks. Anyway the tests seem to run fine locally and the new constructor seems like a good candidate, so code coverage should not be a reason to block this. |
No description provided.