Kpa-clawbot fa5ac2751d fix: debounce distance index rebuild to prevent CPU hot loop ...

On busy meshes (325K+ transmissions, 50 observers), every ingest poll
triggers a full distance index rebuild (1M+ hop records) because
new observations frequently pick longer paths via pickBestObservation.
With 1-second poll intervals, the rebuild never finishes before the
next one starts, pegging CPU at 100% and starving the HTTP server.

Fix: mark the distance index dirty on path changes but only rebuild
at most every 30 seconds. The initial Load() rebuild still runs
synchronously, and distLast is set afterward to prevent an immediate
re-rebuild on the first ingest cycle.

Discovered on Cascadia Mesh instance (cascadiamesh.org) where the
server was completely unresponsive due to continuous distance index
rebuilds consuming all CPU.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

2026-04-03 23:20:49 -07:00

..

ingestor

test: push ingestor coverage from 70% to 84% (#344) (#492)

2026-04-02 17:31:47 -07:00

server

fix: debounce distance index rebuild to prevent CPU hot loop

2026-04-03 23:20:49 -07:00