Changes between Initial Version and Version 1 of Ticket #2412, comment 45


Ignore:
Timestamp:
May 20, 2019 7:26:19 PM (5 months ago)
Author:
jogger
Comment:

Legend:

Unmodified
Added
Removed
Modified
  • Ticket #2412, comment 45

    initial v1  
    77I thereby stand by my OMF and getNextDelay patches in comment 32 or 11/12 (lighter on CPU) for that matter.
    88
    9 Comment 11/12, 32 and 36 all solve the missed peer bug as described. Detailed logging shows the bug is more serious. Even with my sliding send window allocateSendingBytes returns false quite often because of the send window being used up. This means setting the the willsend boolean from an outdated value of _sendWindowBytesRemaining will produce lots of false negatives. These false negatives are also potential missed peers.
     9Comment 11/12, 32 and 36 all solve the missed peer bug as described. Detailed logging shows the bug is more serious. Even with my sliding send window allocateSendingBytes returns false quite often because of the send window being used up. This means setting the willsend boolean from an outdated value of _sendWindowBytesRemaining will produce lots of false negatives. These false negatives are also potential missed peers.
    1010
    1111Therefore my PeerState patch above is an integral part of the solution. However it contains a rare and hard to see integer overflow for slow conns only. Will add a new revision soon.