The first two commits improve observability by making it so that the vmprofile matches the current ingress-drop-flush generation.
The next adds a resetting effect to the ingress drop monitor: if we see a subwindow within the window measured by the ingress drop monitor in which no packets, are dropped, then we assume that the system has healed itself, and we reset our idea of how many packets have dropped.
Finally, we merge https://github.com/raptorjit/raptorjit/pull/260, reducing the incidence of cold-trace compilation. At some point it is no longer important that cold traces get compiled, because the latency of compilation exceeds the latency of interpretation, given that the trace is run extremely infrequently.
This PR includes four patches from Max.
The first two commits improve observability by making it so that the vmprofile matches the current ingress-drop-flush generation.
The next adds a resetting effect to the ingress drop monitor: if we see a subwindow within the window measured by the ingress drop monitor in which no packets, are dropped, then we assume that the system has healed itself, and we reset our idea of how many packets have dropped.
Finally, we merge https://github.com/raptorjit/raptorjit/pull/260, reducing the incidence of cold-trace compilation. At some point it is no longer important that cold traces get compiled, because the latency of compilation exceeds the latency of interpretation, given that the trace is run extremely infrequently.