20260601 sequence triage支持独立时间序列

This commit is contained in:
2026-06-01 23:41:17 +08:00
parent 902f3ba889
commit 9579f65501
4 changed files with 304 additions and 27 deletions

View File

@@ -2,6 +2,7 @@
from __future__ import annotations
import argparse
from concurrent.futures import ThreadPoolExecutor, as_completed
import hashlib
import json
import os
@@ -56,15 +57,20 @@ def rsync_from_remote(target: str, source: str | Path, destination: Path) -> Non
def rsync_run_artifacts_from_remote(target: str, source: str | Path, destination: Path) -> None:
destination.mkdir(parents=True, exist_ok=True)
rsync_base = ["rsync", "-az", "--partial", "--partial-dir=.rsync-partial"]
rsync_base = ["rsync", "-az", "--ignore-missing-args", "--partial", "--partial-dir=.rsync-partial"]
for name in [
"result.ccr",
"result.cir",
"report.json",
"vrps.csv",
"vaps.csv",
"process-time.txt",
"remote-run-meta.json",
"exit-code.txt",
"started-at.txt",
"finished-at.txt",
"stdout.log",
"stderr.log",
]:
run_local([*rsync_base, f"{target}:{source}/{name}", f"{destination}/"])
@@ -260,6 +266,26 @@ def side_config(name: str) -> dict[str, Any]:
raise SystemExit(f"unknown side config: {name}")
def parse_rirs(raw: str) -> list[str]:
rirs = [item.strip() for item in raw.split(",") if item.strip()]
if not rirs:
raise SystemExit("--rirs must contain at least one RIR")
seen: set[str] = set()
invalid: list[str] = []
duplicate: list[str] = []
for rir in rirs:
if rir not in DEFAULT_RIRS:
invalid.append(rir)
if rir in seen:
duplicate.append(rir)
seen.add(rir)
if invalid:
raise SystemExit(f"unsupported RIR(s): {','.join(invalid)}; valid values: {','.join(DEFAULT_RIRS)}")
if duplicate:
raise SystemExit(f"duplicate RIR(s): {','.join(duplicate)}")
return rirs
def build_remote_command(remote_root: Path, side_name: str, side: dict[str, Any], side_label: str, seq: int, rirs: list[str]) -> tuple[Path, str]:
run_dir = remote_root / "experiments" / "sequence" / side_label / f"run_{seq:04d}"
state_dir = remote_root / "experiments" / "sequence" / side_label / "state" / side["rpKind"]
@@ -429,12 +455,57 @@ def run_sequence_triage(local_exp_root: Path, args: argparse.Namespace) -> None:
])
def run_side_sequence(
args: argparse.Namespace,
remote_root: Path,
local_exp_root: Path,
side_label: str,
side_name: str,
side: dict[str, Any],
seq_path: Path,
rirs: list[str],
) -> list[dict[str, Any]]:
side_progress: list[dict[str, Any]] = []
for seq in range(1, args.samples_per_side + 1):
side_progress.append(
run_one_side_sample(args, remote_root, local_exp_root, side_label, side_name, side, seq_path, seq, rirs)
)
return side_progress
def run_one_side_sample(
args: argparse.Namespace,
remote_root: Path,
local_exp_root: Path,
side_label: str,
side_name: str,
side: dict[str, Any],
seq_path: Path,
seq: int,
rirs: list[str],
) -> dict[str, Any]:
rir_label = ",".join(rirs)
print(
f"[run] {side_label} {side_name} seq={seq} rirs={rir_label} schedule={args.schedule_mode}",
flush=True,
)
remote_run_dir = run_remote_sample(args.ssh_target, remote_root, side_name, side, side_label, seq, rirs)
local_run_dir = local_exp_root / side_label / f"run_{seq:04d}"
rsync_run_artifacts_from_remote(args.ssh_target, remote_run_dir, local_run_dir)
item = build_sequence_item(local_exp_root, side_name, side_label, side, seq, local_run_dir)
item["scheduleMode"] = args.schedule_mode
append_jsonl(seq_path, item)
print(
f"[done] {side_label} seq={seq} wallMs={item.get('wallMs')} vrps={item.get('vrps')} vaps={item.get('vaps')} objects={item.get('cirObjectCount')} rejects={item.get('cirRejectCount')}",
flush=True,
)
return item
def run_experiment(args: argparse.Namespace) -> None:
if not args.skip_build:
build_tool_binaries()
rirs = [item.strip() for item in args.rirs.split(",") if item.strip()]
if set(rirs) != set(DEFAULT_RIRS) or len(rirs) != len(DEFAULT_RIRS):
raise SystemExit("#043 remote acceptance requires all5 RIRs: afrinic,apnic,arin,lacnic,ripe")
rirs = parse_rirs(args.rirs)
left = side_config(args.left)
right = side_config(args.right)
run_root = Path(args.run_root).resolve()
@@ -447,6 +518,7 @@ def run_experiment(args: argparse.Namespace) -> None:
"right": args.right,
"samplesPerSide": args.samples_per_side,
"rirs": rirs,
"scheduleMode": args.schedule_mode,
"remoteRoot": str(remote_root),
"sshTarget": args.ssh_target,
})
@@ -467,19 +539,24 @@ def run_experiment(args: argparse.Namespace) -> None:
left_seq_path.unlink(missing_ok=True)
right_seq_path.unlink(missing_ok=True)
progress: list[dict[str, Any]] = []
for seq in range(1, args.samples_per_side + 1):
for side_label, side_name, side, seq_path in [("A", args.left, left, left_seq_path), ("B", args.right, right, right_seq_path)]:
print(f"[run] {side_label} {side_name} seq={seq} all5", flush=True)
remote_run_dir = run_remote_sample(args.ssh_target, remote_root, side_name, side, side_label, seq, rirs)
local_run_dir = local_exp_root / side_label / f"run_{seq:04d}"
rsync_run_artifacts_from_remote(args.ssh_target, remote_run_dir, local_run_dir)
item = build_sequence_item(local_exp_root, side_name, side_label, side, seq, local_run_dir)
append_jsonl(seq_path, item)
progress.append(item)
print(
f"[done] {side_label} seq={seq} wallMs={item.get('wallMs')} vrps={item.get('vrps')} vaps={item.get('vaps')} objects={item.get('cirObjectCount')} rejects={item.get('cirRejectCount')}",
flush=True,
)
if args.schedule_mode == "interleaved":
for seq in range(1, args.samples_per_side + 1):
for side_label, side_name, side, seq_path in [
("A", args.left, left, left_seq_path),
("B", args.right, right, right_seq_path),
]:
progress.append(
run_one_side_sample(args, remote_root, local_exp_root, side_label, side_name, side, seq_path, seq, rirs)
)
else:
with ThreadPoolExecutor(max_workers=2) as executor:
futures = [
executor.submit(run_side_sequence, args, remote_root, local_exp_root, "A", args.left, left, left_seq_path, rirs),
executor.submit(run_side_sequence, args, remote_root, local_exp_root, "B", args.right, right, right_seq_path, rirs),
]
for future in as_completed(futures):
progress.extend(future.result())
progress.sort(key=lambda item: (str(item.get("side")), int(item.get("seq") or 0)))
write_json(local_exp_root / "run-progress.json", progress)
run_sequence_triage(local_exp_root, args)
ssh_script(args.ssh_target, f"df -h /data / > {shlex.quote(str(remote_root / 'df-after.txt'))} 2>&1 || true; free -h > {shlex.quote(str(remote_root / 'free-after.txt'))} 2>&1 || true")
@@ -496,6 +573,7 @@ def main() -> None:
parser.add_argument("--right", default="rpki-client-standard")
parser.add_argument("--samples-per-side", type=int, default=3)
parser.add_argument("--rirs", default=",".join(DEFAULT_RIRS))
parser.add_argument("--schedule-mode", choices=["interleaved", "parallel"], default="interleaved")
parser.add_argument("--align-window-runs", type=int, default=2)
parser.add_argument("--align-window-secs", type=int, default=1800)
parser.add_argument("--sample-limit", type=int, default=200)