forked from mtusi2026/mtusi2026lab4
lab4: память загрузок и S3 под лимитом, документация защиты
- тело загрузки: один буфер ровно Content-Length, без копий до S3 и djpeg; - PNG: zlib получает вход кусками по 64 KiB (unconsumed_tail копировал остаток IDAT); - djpeg, убитый OOM-killer'ом, -> 503 (ресурсы сервера), а не 415; - api: memswap_limit = memory (жёсткий лимит без swap); S3: GOMEMLIMIT и лимиты одновременных загрузок SeaweedFS, память 1g; - docs: CONSISTENCY.md, DEFENSE.md, WINDOWS.md, DATA_MODEL.md под lab4; scripts/demo_attachments.py для сценария защиты; CI и collect_evidence под lab4; - bench/reconcile_faults.py: отчёты трека 2 (ошибки с обеих сторон, прерванный apply). Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
co-authored by
Claude Opus 5.5
parent
a2d867bec6
commit
4ea6924db1
+36
-7
@@ -57,6 +57,14 @@ def cg(cid: str, name: str) -> int:
|
||||
return int(sh("docker", "exec", cid, "cat", f"/sys/fs/cgroup/{name}").split()[0])
|
||||
|
||||
|
||||
def memory_stat(cid: str, key: str) -> int:
|
||||
for line in sh("docker", "exec", cid, "cat", "/sys/fs/cgroup/memory.stat").splitlines():
|
||||
name, value = line.split()
|
||||
if name == key:
|
||||
return int(value)
|
||||
return 0
|
||||
|
||||
|
||||
def vmhwm(cid: str) -> int:
|
||||
for line in sh("docker", "exec", cid, "cat", "/proc/1/status").splitlines():
|
||||
if line.startswith("VmHWM:"):
|
||||
@@ -64,6 +72,16 @@ def vmhwm(cid: str) -> int:
|
||||
return 0
|
||||
|
||||
|
||||
def container_events() -> dict[str, str]:
|
||||
"""restart count + OOM-kill events per container (cgroup memory.events oom_kill counter)."""
|
||||
out = {}
|
||||
for cid in sh("docker", "compose", "ps", "-q").split():
|
||||
name, restarts = sh("docker", "inspect", "-f", "{{.Name}} {{.RestartCount}}", cid).split()
|
||||
events = dict(line.split() for line in sh("docker", "exec", cid, "cat", "/sys/fs/cgroup/memory.events").splitlines())
|
||||
out[name.lstrip("/")] = f"restarts={restarts} oom_kill={events.get('oom_kill', '?')}"
|
||||
return out
|
||||
|
||||
|
||||
def recreate(env: dict) -> None:
|
||||
sh("docker", "compose", "up", "-d", "--force-recreate", "--no-deps", "--wait", "api", env=env, timeout=300)
|
||||
sh(sys.executable, "scripts/lbctl.py", "wait-up", str(len(replicas())), "--timeout", "90") # LB sees them UP
|
||||
@@ -129,7 +147,9 @@ def run_scenario(name: str, env: dict, files: list[tuple[bytes, str]]) -> dict:
|
||||
results[i] = (r.status_code, time.perf_counter() - started, r.headers.get("x-instance-id", "-"))
|
||||
c.close()
|
||||
|
||||
samplers = [Sampler(cid) for cid in ids]
|
||||
s3_id = sh("docker", "compose", "ps", "-q", "s3").strip()
|
||||
restarts_before = container_events()
|
||||
samplers = [Sampler(cid) for cid in ids] + [Sampler(s3_id)]
|
||||
for s in samplers:
|
||||
s.start()
|
||||
time.sleep(0.5)
|
||||
@@ -144,7 +164,7 @@ def run_scenario(name: str, env: dict, files: list[tuple[bytes, str]]) -> dict:
|
||||
for s in samplers:
|
||||
s.stop()
|
||||
per_replica = {}
|
||||
for cid, s in zip(ids, samplers):
|
||||
for cid, s in zip(ids, samplers[:-1]):
|
||||
hostname = sh("docker", "inspect", "-f", "{{.Config.Hostname}}", cid).strip()
|
||||
per_replica[hostname] = {
|
||||
"idle_mib": round(idle[cid] / MiB, 1),
|
||||
@@ -162,6 +182,11 @@ def run_scenario(name: str, env: dict, files: list[tuple[bytes, str]]) -> dict:
|
||||
codes[str(r[0])] = codes.get(str(r[0]), 0) + 1
|
||||
summary = {
|
||||
"scenario": name, "settings": env, "clients": len(files),
|
||||
"s3_sampled_max_mib": round(max(samplers[-1].values or [0]) / MiB, 1),
|
||||
"s3_anon_after_mib": round(memory_stat(s3_id, "anon") / MiB, 1),
|
||||
"s3_limit_mib": round(cg(s3_id, "memory.max") / MiB, 1),
|
||||
# OOM kills / restarts DURING the scenario (any container of the stand): must be empty
|
||||
"incidents": {k: v for k, v in container_events().items() if v != restarts_before.get(k)},
|
||||
"file_mib_each": round(statistics.mean(len(d) for d, _ in files) / MiB, 2),
|
||||
"types": sorted({t for _, t in files}), "status_codes": codes, "wall_seconds": round(wall, 2),
|
||||
"latency_s": {"p50": round(lat[len(lat) // 2], 2), "max": round(lat[-1], 2)} if lat else None,
|
||||
@@ -179,16 +204,20 @@ def main() -> int:
|
||||
out = ROOT / args.output
|
||||
out.mkdir(parents=True, exist_ok=True)
|
||||
print("generating files ...", flush=True)
|
||||
pngs = [(imagegen.png_of_size(int(9.5 * MiB)), "image/png") for _ in range(args.clients)]
|
||||
pngs = [(imagegen.png_of_size(int(9.5 * MiB)), "image/png") for _ in range(2 * args.clients)]
|
||||
jpegs = [(big_jpeg(8000, 5000, i), "image/jpeg") for i in range(args.clients)]
|
||||
base = {"UPLOAD_CONCURRENCY": "4", "IMAGE_VERIFY_CONCURRENCY": "1"}
|
||||
n, n2 = args.clients, 2 * args.clients
|
||||
scenarios = [
|
||||
("png-9.5MiB x16, slots=4 (default)", base, pngs),
|
||||
("png-9.5MiB x16, slots=16 (no bound)", {"UPLOAD_CONCURRENCY": "16", "IMAGE_VERIFY_CONCURRENCY": "1"}, pngs),
|
||||
("jpeg-40Mpx-progressive x16, slots=4, verify=1 (default)", base, jpegs),
|
||||
("jpeg-40Mpx-progressive x16, slots=4, verify=4", {"UPLOAD_CONCURRENCY": "4", "IMAGE_VERIFY_CONCURRENCY": "4"}, jpegs),
|
||||
(f"png-9.4MiB x{n}, slots=4 (default)", base, pngs[:n]),
|
||||
(f"png-9.4MiB x{n2}, slots=4 (default)", base, pngs),
|
||||
(f"png-9.4MiB x{n2}, slots=16 (bound raised)", {"UPLOAD_CONCURRENCY": "16", "IMAGE_VERIFY_CONCURRENCY": "1"}, pngs),
|
||||
(f"jpeg-40Mpx-progressive-4:4:4 x{n}, verify=1 (default)", base, jpegs),
|
||||
(f"jpeg-40Mpx-progressive-4:4:4 x{n}, verify=4", {"UPLOAD_CONCURRENCY": "4", "IMAGE_VERIFY_CONCURRENCY": "4"}, jpegs),
|
||||
]
|
||||
results = []
|
||||
sh("docker", "compose", "restart", "s3") # S3 memory counters start from an idle process
|
||||
sh("docker", "compose", "up", "-d", "--wait", "s3")
|
||||
try:
|
||||
for name, env, files in scenarios:
|
||||
results.append(run_scenario(name, env, files))
|
||||
|
||||
Reference in New Issue
Block a user