hpc-orchestration

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Chinese

Use this skill as the repository-level execution layer above solver-specific HPC skills.

将此技能用作求解器专属HPC技能之上的仓库级执行层。

Read
```
references/cluster-operations-manual.md
```
first for any end-to-end cluster workflow.
Read
```
references/lifecycle-manual.md
```
when coordinating solver skills with orchestration stages.
Read
```
references/scheduler-and-parallelism.md
```
when choosing scheduler directives or MPI sizing.
Read
```
references/slurm-launch-patterns.md
```
when the main decision is how to launch work inside a Slurm allocation.
Read
```
references/environment-and-storage-hygiene.md
```
when environment modules, scratch policy, or filesystem behavior may decide workflow reliability.
Read
```
references/data-transfer-and-staging.md
```
when files must be synchronized, staged, archived, or verified.
Read
```
references/software-build-and-reproducibility.md
```
when compilation, package stacks, or rebuildability are in scope.
Read
```
references/interactive-debugging-and-profiling.md
```
when the task needs live diagnosis, scaling studies, or profiler evidence.
Read
```
references/remote-development-and-notebooks.md
```
when VS Code Remote SSH, Jupyter, or port forwarding is involved.
Read
```
references/container-workflows.md
```
when Apptainer or Singularity-style execution is in scope.
Read
```
references/public-protocol.md
```
when aligning solver skills with the repository lifecycle contract.
Read
```
references/error-pattern-dictionary.md
```
when scheduler, monitoring, or log-tracking failures occur.

对于任何端到端集群工作流，请先阅读
```
references/cluster-operations-manual.md
```
。
当协调求解器技能与编排阶段时，请阅读
```
references/lifecycle-manual.md
```
。
当选择调度器指令或MPI规模时，请阅读
```
references/scheduler-and-parallelism.md
```
。
当主要决策是如何在Slurm分配内启动工作时，请阅读
```
references/slurm-launch-patterns.md
```
。
当环境模块、临时存储策略或文件系统行为可能影响工作流可靠性时，请阅读
```
references/environment-and-storage-hygiene.md
```
。
当需要同步、暂存、归档或验证文件时，请阅读
```
references/data-transfer-and-staging.md
```
。
当涉及编译、包栈或可重建性时，请阅读
```
references/software-build-and-reproducibility.md
```
。
当任务需要实时诊断、扩展研究或性能分析器证据时，请阅读
```
references/interactive-debugging-and-profiling.md
```
。
当涉及VS Code Remote SSH、Jupyter或端口转发时，请阅读
```
references/remote-development-and-notebooks.md
```
。
当涉及Apptainer或Singularity风格的执行时，请阅读
```
references/container-workflows.md
```
。
当使求解器技能与仓库生命周期契约保持一致时，请阅读
```
references/public-protocol.md
```
。
当出现调度器、监控或日志跟踪故障时，请阅读
```
references/error-pattern-dictionary.md
```
。

Load these on demand:

```
references/tools-and-scripts.md
```
for shared execution tool roles
```
references/ecosystem-roadmap.md
```
for repository-wide coverage goals
```
references/legacy-template.md
```
when adapting an older solver skill draft into the current format

按需加载以下内容：

Use

assets/templates/

when a concrete scheduler scaffold is needed:

当需要具体的调度器脚手架时，请使用

assets/templates/

中的内容：

Use

scripts/

for deterministic orchestration tasks:

对于确定性编排任务，请使用

scripts/

中的内容：

Do not run heavy MPI workloads on login nodes.
Do not separate queue submission from runtime monitoring in a production workflow.
Do not resubmit unchanged failing jobs when a solver-specific error dictionary exists.
Do not choose core counts without a scale-based heuristic.