P2P: Misc improvements

[heifner]

In preparation of work for #452 some initial cleanup.

Testing with performance harness when sending more transactions than nodeos can process it fills up the write queue. This is designed to then close the connection. nodeos was generating thousands of write queue full messages (example below) because the posted close() call would have to wait until all the currently queued up tasks on the strand finished. This delayed the close for seconds. Added a closing flag to stop existing tasks quickly, resulting in only one message instead of thousands and an immediate close.

error 2023-04-15T16:46:16.320 net-0     net_plugin.cpp:2494           start_read_message   ] ["localhost:9877 - d7c63c8" - 1 127.0.0.1:37658] write queue full 71936405 bytes, giving up on connection, closing

Added better encapsulation of the sync_mtx.
Added alignas(hardware_destructive_interference_size) to avoid cache synchronization.

[greg7mdp]

Just curious. What led you to adding alignas(hardware_destructive_interference_size) to avoid cache synchronization.?
Could you measure whether it made a difference in performance?

[heifner]

Just curious. What led you to adding alignas(hardware_destructive_interference_size) to avoid cache synchronization.? Could you measure whether it made a difference in performance?

Just something I've been meaning to add. No, I have not measured the performance difference.

[greg7mdp]

Just something I've been meaning to add. No, I have not measured the performance difference.

Very cool change!