Telemetry and resource settings
VM Settings → Telemetry transport… configures how syscall records move from the guest to the host. The configurable transport needs a base made for Antinomie 0.3.0.
Settings
Section titled “Settings”| Setting | Default | Effect |
|---|---|---|
| Guest buffer per vCPU | 1 MiB | Select 256 KiB to 4 MiB per virtual CPU to absorb bursts |
| When buffers fill | Never drop | Wait for host capacity instead of immediately dropping a new record |
| Longest wait per event | 5 seconds | A record can still be dropped if the wait deadline expires |
| Host event table | 16,384 calls | Capacity for calls the host is assembling |
The Drop new telemetry choice keeps the guest from waiting for buffer capacity, with explicit loss when full.
Interpret Never drop correctly
Section titled “Interpret Never drop correctly”The name describes the waiting policy, not an unconditional guarantee. The configured wait timeout, capture errors, recording limits and shutdown can still leave an observation incomplete.
Waiting changes the observed program’s timing. A disk or host that cannot keep up can slow the guest. The dropping mode avoids that buffer wait but can lose events. Choose according to the measurement you need.
Older 0.2.0 and 0.1.0 bases still run with their original 256 KiB buffers and drop-on-full behavior. They do not gain the new policy by updating only the app.
Host and guest memory
Section titled “Host and guest memory”The per-vCPU buffer is guest memory in addition to the guest’s ordinary working set. Four CPUs at the default setting allocate 4 MiB for those buffers. Increasing guest buffers does not replace host event-table or storage capacity.
Recording writes use a background writer with a bounded 256 MiB queue. The host also needs memory for the desktop, live tables and analysis. Leave headroom instead of assigning all host RAM to a VM.
Watch the counters
Section titled “Watch the counters”Read transport, UI/event-table and storage loss as separate conditions:
- Transport loss means some guest observations did not reach the host.
- UI/event-table loss affects the host’s assembled live view.
- Storage loss or quota state affects retained evidence.
0.3.0 retires completed calls from the host table first. Increasing table size can help concurrent unfinished calls, but cannot correct missing guest data.
If loss grows, narrow the trace, check host CPU/disk pressure, and consider a larger buffer or waiting policy. Recheck the counters after changing one setting.