Skip to content

kali20gakki/msAgent / skills/host-cpu/msagent-perf-cpu-bind-tune

msagent-perf-cpu-bind-tune

Flagged

昇腾 NPU 多卡任务的 Host CPU 绑核方案求解器:覆盖 PyTorch 训练/离线推理与 vLLM-Ascend、SGLang 推理服务的 rank/worker/engine 分核,基于 NUMA 拓扑与 卡↔进程映射输出可执行、可回滚、可验证的绑核计划(taskset 前缀 / cpuset / OMP 环境变量),自动判定事前(env 片段)与事中(在线 sched_setaffinity)双形态。 触发:绑核方案、核怎么分、per-rank 绑核、NUMA 绑定方案、cpuset 划分、多卡分核、 CPU range 冲突、TTFT 或吞吐受 CPU 影响。 不用于:绑核没生效/绑核异常诊断(→ perf-cpu-config-audit);运行期是否落核 (→ perf-cpu-tracing);变慢根因归因(→ perf-cpu-perf-triage)。

scripts/zone_alloc.py

View on GitHub
# -------------------------------------------------------------------------
# This file is part of the MindStudio project.
# Copyright (c) 2026 Huawei Technologies Co.,Ltd.
#
# MindStudio is licensed under Mulan PSL v2.
# You can use this software according to the terms and conditions of the Mulan PSL v2.
# You may obtain a copy of Mulan PSL v2 at:
#
#          http://license.coscl.org.cn/MulanPSL2
#
# THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND,
# EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT,
# MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE.
# See the Mulan PSL v2 for more details.
# -------------------------------------------------------------------------

"""划区与核预算:每个 node 预留系统中断区/干扰隔离区,其余为业务区。

A2 实机:24 CPU/node、无 SMT;默认 system 2 + interference 1 = 3,
业务区 21;同一 node 上多个 unit 均分连续核集。
"""

import topology


DEFAULT_POLICY = {
    "smt": "exclusive",
    "irq_reserve_per_node": 2,
    "interference_reserve_per_node": 1,
}


def allocate(topo, units, policy=None):
    policy = dict(DEFAULT_POLICY, **(policy or {}))
    by_node = {}
    for unit in units:
        node = unit.get("target_node")
        if node is None:
            continue
        by_node.setdefault(node, []).append(unit)

    zones = {}
    alloc = {}
    for node in topo.get("nodes") or []:
        nid = node["node"]
        node_units = by_node.get(nid, [])
        if node_units:
            cpus = sorted({
                c for u in node_units for c in (u.get("target_cpus") or [])
            })
            if not cpus:
                cpus = list(node["cpus"])
        else:
            cpus = list(node["cpus"])

        irq_n = int(policy.get("irq_reserve_per_node") or 0)
        int_n = int(policy.get("interference_reserve_per_node") or 0)
        # 预留区从 CPU 编号高端截取(尾部),业务区保持低端连续:
        # 1) unit 核集(compute/comm/io/main)始终连续,便于 OMP_PLACES=cores;
        # 2) 内核 softirq/irqbalance 默认倾向低端 CPU,高端核天然受打扰概率更低;
        # 3) 预留大于 node 半数时放弃预留(小 node 全给业务),由 C1 报告预算问题。
        if irq_n + int_n >= len(cpus):
            irq_n, int_n = 0, 0
        interference = cpus[-int_n:] if int_n else []
        system = cpus[-(int_n + irq_n):(-int_n if int_n else None)] if irq_n else []
        business = cpus[:-(int_n + irq_n)] if (int_n + irq_n) else cpus

        # 只保留 cpulist 一份核集描述,避免 cpus/cpulist 双字段口径漂移
        zones[str(nid)] = {
            "node": nid,
            "cpulist": topology.cpus_to_str(cpus),
            "system_irq": topology.cpus_to_str(system),
            "interference": topology.cpus_to_str(interference),
            "business": topology.cpus_to_str(business),
            "business_cpus": business,
            "unit_ids": [u["unit_id"] for u in node_units],
        }

        if node_units:
            for unit, cpuset in zip(
                sorted(node_units, key=lambda x: x["unit_id"]),
                _split_contiguous(business, len(node_units))
            ):
                alloc[unit["unit_id"]] = cpuset

    return {
        "policy": policy,
        "zones": zones,
        "alloc": alloc,
    }


def _split_contiguous(cpus, n):
    if n <= 0:
        return []
    if n == 1:
        return [list(cpus)]
    base = len(cpus) // n
    rem = len(cpus) % n
    out = []
    idx = 0
    for i in range(n):
        size = base + (1 if i < rem else 0)
        out.append(list(cpus[idx:idx + size]))
        idx += size
    return out
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 471kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall