[IGNITE-13016] Fix backward checking of failed node.

• Type: Sub-task
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Fix Version/s: 2.9
• Labels: iep-45
• Assignee: Vladimir Steshin
• Reporter: Vladimir Steshin
• Reviewer: Sergey Chugunov
• Created: 15/May/20 13:28
• Updated: 21/Jul/20 14:38
• Resolved: 21/Jul/20 10:35

Description

Backward node connection checking looks wierd. What we might improve are:

1) Addresses checking could be done in parrallel, not sequentially.

for (InetSocketAddress addr : nodeAddrs) {
    // Connection refused may be got if node doesn't listen
    // (or blocked by firewall, but anyway assume it is dead).
    if (!isConnectionRefused(addr)) {
        liveAddr = addr;

        break;
        }
}

2) Any io-exception should be considered as failed connection, not only connection-refused:

catch (ConnectException e) {
    return true;
}
catch (IOException e) {
    return false;
}

3) Timeout on connection checking should not be constant or hardcode:

sock.connect(addr, 100);

4) Decision to check connection should rely on configured exchange timeout, no on the ping interval

// We got message from previous in less than double connection check interval.
boolean ok = rcvdTime + U.millisToNanos(connCheckInterval) * 2 >= now;

Issue Links

• causes: IGNITE-13111 Simplify backward checking of node connection.
• depends upon: IGNITE-13012 Fix failure detection timeout. Simplify node ping routine.
• incorporates: IGNITE-13017 Remove hardcoded delay from re-marking failed node as alive.
• is depended upon by: IGNITE-13014 Remove double checking of node availability.
• links to: GitHub Pull Request #7838