sender queue, not on the way out, so that SSU requests
bandwidth allocations for each packet in parallel
and competes more effectively with NTCP for bandwidth.
Inbound stubbed-out only.
so I/O doesn't stall under high CPU load
- Raise DH generator thread priority to keep
DH building out of event pumper thread
- Raise PRNG and YK generator thread priorites one notch
- Set I2PThread priority in constructor
Fixes problems mainly seen on Windows, which seems
to be much more sensitive to priority settings
at least one OB tunnel, so the client waits until we are ready.
This will reduce drops, retransmissions, and failures on new client tunnels.
Fixes to prevent multiple pending LS requests.
- When socket is closed, set sessionID and LS to null,
close subsession and set its sessionID and LS to null
- Checks on client side for null session ID
- Check for null session in Destroy Session message
- Don't kill I2CP connection due to a bad session ID
in a SendMessage, just drop the message and send
a MessageStatusMessage
- Log tweaks
Catch null SessionId in messages and stop connection instead of NPE
Wait for LS in SubSession in connect() so we don't send data w/o
a session ID and leaseset
Log tweaks
Send our own RI unsolicited in reply if we aren't floodfill
Don't ack or flood a store of an unknown type
PeerTestJob: Don't generate zero reply token
Tunnels: More checks of messages received down exploratory tunnels
javadocs and comments
- Directly connect to nearby floodfills to share our RI to speed up integration of new floodfills (#1195).
- Called on both non-ff -> ff OR ff -> non-ff.
- Create FloodfillRouterInfoFloodJob to do so.
