feat(kvb-org-organize): hold heavy jobs back on memory pressure with one shipped default config

watch.py admits heavy work only while PSI, swap-out rate and RAM forecast stay under the configured limits; config.default.json is the single default the install writes once (mode 600, no clobber) and every hook, admit and queue command falls back to. Browser runners and package-manager script forms count as heavy, preflight derives its probes from the light and heavy tool lists, and the shipped default carries no prune command, buildx path or host-specific key. Tests run once on unify/skills-20261114 after the merge (operator order 2026-11-14).

🎯 Fix: 85 %
💥 Break: 8 %
💸 Tokens: 99 %
⏱️ Velocity: –
🛡️ Security: 100 %
🏆 Score: 79
🚦 Gate: ⛔ fix 85 % < 90, break 8 % > 5

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Kevin Veen-BirkenbachandClaude Opus 5.5 committed 2026-11-14 21:17:28 +00:00
1 parent a2a03995cb
commit 8e2db852e5
27 files changed
+2795 -207

No files matched your search

+52 -7
View File
@@ -2,12 +2,17 @@
"""Args:
sample: append one row to <state>/samples.jsonl: {t, disks: [{mount,
used, total}], cpu, ram, swap, load, gpus: [{name, util, vram_used,
vram_total} | {name, asleep: true}], queue: {waiting, active}}; t in
epoch seconds, used, total and vram in bytes, cpu, ram, swap and util
in percent, load the 1-minute load average, a GPU named by its PCI
address. cpu is the busy share of the /proc/stat ticks since those
the run before kept in <state>/cpu.json as [t, idle, total], None on
the first run and after a gap of more than 15 minutes. An NVIDIA GPU
vram_total} | {name, asleep: true}], queue: {waiting, active},
psi_some, psi_full, swap_out_mb_min}; t in epoch seconds, used, total
and vram in bytes, cpu, ram, swap and util in percent, load the
1-minute load average, a GPU named by its PCI address. cpu is the
busy share of the /proc/stat ticks since those the run before kept in
<state>/cpu.json as [t, idle, total], None on the first run and after
a gap of more than 15 minutes. psi_some and psi_full are the avg10 of
/proc/pressure/memory, None without PSI; swap_out_mb_min the MB per
minute swapped out (pswpout of /proc/vmstat) since the count the run
before kept in <state>/swap.json as [t, pswpout], None like cpu and
after a counter reset. An NVIDIA GPU
whose PCI runtime_status is not active is recorded asleep, and
nvidia-smi is not run for it; an awake one nvidia-smi gives no
reading for (no driver, a driver mismatch) keeps its name with util
@@ -75,12 +80,15 @@ NVIDIA_QUERY = "pci.bus_id,utilization.gpu,memory.used,memory.total"
UNREAD = {"util": None, "vram_used": None, "vram_total": None}
COUNTS = (("input", "input_tokens"), ("output", "output_tokens"), ("cache_read", "cache_read_input_tokens"), ("cache_write", "cache_creation_input_tokens"))
NAMES = tuple(name for name, _ in COUNTS)
HOST_DIGITS = (("cpu", 1), ("ram", 1), ("swap", 1), ("load", 2))
HOST_DIGITS = (("cpu", 1), ("ram", 1), ("swap", 1), ("load", 2), ("psi_some", 2), ("psi_full", 2), ("swap_out_mb_min", 1))
CLAIMS = ("*/claim.json", "*/*/claim.json")
SAMPLES = "samples.jsonl"
TOKENS = "tokens.jsonl"
OFFSETS = "offsets.json"
TICKS = "cpu.json"
SWAPPED = "swap.json"
PAGE_BYTES = os.sysconf("SC_PAGE_SIZE")
MB = 1_000_000
def read_text(path: Path) -> str:
@@ -263,9 +271,43 @@ def cpu_since(proc: Path, state: Path, t: int) -> float | None:
return cpu_percent(before, ticks)
def psi_avg10(text: str, kind: str) -> float | None:
for line in text.splitlines():
fields = line.split()
if fields[:1] == [kind]:
values = dict(field.split("=", 1) for field in fields[1:] if "=" in field)
return number(values.get("avg10", ""))
return None
def swapped_pages(text: str) -> int | None:
for line in text.splitlines():
fields = line.split()
if len(fields) == 2 and fields[0] == "pswpout" and fields[1].isdigit():
return int(fields[1])
return None
def swap_out_since(proc: Path, state: Path, t: int) -> float | None:
"""Returns:
The MB per minute swapped out since the pswpout count the run before
kept in <state>/swap.json at most BUCKET seconds before t, else None,
also after a counter reset; this run's count replaces it.
"""
pages = swapped_pages(read_text(proc / "vmstat"))
if pages is None:
return None
stored = read_json(state / SWAPPED)
replace(state / SWAPPED, json.dumps([t, pages]))
if not integers(stored, 2) or not t - BUCKET <= stored[0] < t or stored[1] > pages:
return None
return round((pages - stored[1]) * PAGE_BYTES / MB * 60 / (t - stored[0]), 1)
def sample(proc: Path, sys_root: Path, council: Path, run: Callable[[list[str]], str | None], state: Path) -> dict:
t = int(time.time())
ram, swap = memory(parse_meminfo(read_text(proc / "meminfo")))
pressure = read_text(proc / "pressure" / "memory")
return {
"t": t,
"disks": parse_df(run(["df", "-P", "-k", "-l"]) or ""),
@@ -275,6 +317,9 @@ def sample(proc: Path, sys_root: Path, council: Path, run: Callable[[list[str]],
"load": parse_loadavg(read_text(proc / "loadavg")),
"gpus": gpus(sys_root, run),
"queue": queue(council),
"psi_some": psi_avg10(pressure, "some"),
"psi_full": psi_avg10(pressure, "full"),
"swap_out_mb_min": swap_out_since(proc, state, t),
}