UI Stability, Memory Management and More - B3 Work List

[thebracket]

(Please squash merge this when ready).

• Fix “unknown” flags in the UI.
• Fix “download/download” on the tree.
• Fix “break system packages” flag on Debian installer, removing the “Python requirements.txt not installed” problems on older versions of Ubuntu.
• Fix a typo that prevent Spylnx integrations from working automatically in the scheduler.
• The title of your shaper node appears properly in your browser title bar now.
• The webserver no longer freezes.
• “Generated Site” no longer fills up the Sankeys.
• The main tree Sankey uses a ring buffer to smooth rendering (this is coming to other elements in the next beta).
• Nicer Sankey ribbon colors, especially in dark mode.
• Improved margins around several graphs to make the axis labels visible.
• The tick thread is now very careful to never overrun its desired time period; time-based spikes on usage (which spread to Long-Term Stats) are a thing of the past.
• You can add disable_webserver = false to your /etc/lqos.conf if you don’t want to have a web GUI at all.
• Filtered out “one way” flows from the ASN explorer (these are usually bot scans, and will be tracked separately in the next beta).
• Internal code:
  • Fix MANY warnings.
  • MANY diagnostic messages moved to “debug” level so they aren’t visible by default.
  • Transition from “log” to “tracing” for log handling. Improves performance, and is the new de-facto standard in Rust.
  • Reduced dependency on DashMap.
  • Replace channels with Crossbeam for greater performance and improved efficiency.
  • All threads are now named, you can watch them individually in htop!
  • Updated several dependency versions.
• Memory usage:
  • eBPF now has periodic garbage collection checks, expiring ancient history from the hosts list.
  • Configuration management is now much more memory-efficient.
  • Queue structure is stored more efficiently.
  • Shaped devices is stored more efficiently.
  • Switch to the mimalloc allocator for reduced total memory usage.
  • Flow tracking is significantly more efficient.
  • All channels are now bounded.
  • The webserver is now contained to a single thread.
  • Network.json handling is significantly more efficient.
  • MANY hot-loop allocations have been moved out of the hot loop and into a pre-made buffer.
  • Reduced size of preallocated buffers; it can grow if needed.