It looks like in production we backoff far too aggressively.
This is caused by having multiple concurrent tryConnect() attempts
which will all "succeed" or "fail" and thus apply the doubling
logic many times ballooning the retry attempt delay.
I've also added a hard gap of 30s backoff so every 30s we will
try to open the connection.
It looks like in production we backoff far too aggressively.
This is caused by having multiple concurrent
tryConnect()
attempts which will all "succeed" or "fail" and thus apply the doubling logic many times ballooning the retry attempt delay.I've also added a hard gap of 30s backoff so every 30s we will try to open the connection.
r: @syyang