[torlando-tech]

Problem

The iPhone 14 *.diskwrites_resource .ips reports (limit of 12,137 KB over 86,400 s) are what get the background transport throttled and killed — and the Memory Usage screen had no way to show disk writes, because the NE memory trace carried no disk counter.

Change

• Trace layout v2 (record 64 → 72 B): every record now carries the cumulative proc_pid_rusage → ri_diskio_byteswritten at sample time, sampled by the existing 1 Hz loop alongside the footprint. Boot/stop markers carry it too, so kill cards can show a dead generation's total.
  • proc_pid_rusage is not declared in the iOS SDK headers (libproc.h is macOS-only); resolved once via dlsym(RTLD_DEFAULT, …) — reading your own rusage needs no entitlement.
  • ensureFile now drops and recreates a file whose size does not match this build's layout (a v1 ring from an older build) instead of interleaving two geometries. The trace is diagnostics, not state — no migration.
• Analyzer: per-point write rates (delta between consecutive plotted records within a generation — no live/bucket double counting) and per-event disk totals (last − first of the cumulative counter; nil for pre-v2 rings).
• Screen: new "Disk writes" card — cumulative MB per process run with the ~11.9 MB/day budget dashed line (12,137 KB from the .ips reports), plus a disk total line on the kill card.

Testing

• NEMemoryTraceTests: 24/24 on the Mac sim (layout round-trip incl. the new field, v1-file drop+recreate, kill disk total + per-point rate, counter reset across generations).
• Full ColumbaAppTests suite running; device verification of the v2 ring + chart pending a device build.

[greptile-apps]

Confidence Score: 5/5

[Medium impact] The change since the previous review appears safe to merge.

Summary

The PR adds disk-write counters to the Network Extension’s memory trace and shows write rates against the daily budget.

• The shared trace records disk writes for each Network Extension run.
• The Memory Usage screen shows disk-write rates and run totals.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Network Extension samples disk writes] --> B[Trace layout v2]
  B --> C[Analyzer computes rates and totals]
  C --> D[Memory screen shows disk writes]

Loading

_{Reviews (7) · Last reviewed commit: "DiskIO: Darwin's RTLD_DEFAULT is (void*)..." · Reviewed by Greptile}

[torlando-tech]

@greptile review

[torlando-tech]

All three P2s fixed in f04979e (+ chart nil-rate skip):

1. Run totals lose older writes — the counter is cumulative since process start, so the kill total is now the LAST reading outright; the last − first subtraction is gone, which also removes the ring-wrapped undercount. Regression test testWrappedRingDoesNotDropEarlyWritesFromTheKillTotal pins the 19 MB early-writes case.
2. Failed samples look like zero — a failed probe now stores the diskWrittenUnavailable sentinel (never 0); the analyzer renders it as no data, gives that point and the next no rate, and a generation ending on it reports no total. testFailedProbeIsNoDataRatherThanZero.
3. Healthy runs look over budget — the card now plots the write rate (KB/s, 60 s-window averages) against the daily budget as a rate (~0.14 KB/s); bursts above it read as bursts, not breaches.

NEMemoryTraceTests 26/26 on the Mac sim.

[torlando-tech]

@greptile review

[torlando-tech]

@greptile review

[torlando-tech]

Fixed: the once-only log is now guarded by a dedicated NSLock (check-and-set inside, log call outside).

[torlando-tech]

@greptile review

[torlando-tech]

@greptile review