runops 2.3.0__tar.gz → 2.3.1__tar.gz

{runops-2.3.0 → runops-2.3.1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.5
 Name: runops
-Version: 2.3.0
+Version: 2.3.1
 Summary: Run ledger and Slurm job submission for HPC simulation projects
 Project-URL: Homepage, https://github.com/Nkzono99/runops
 Project-URL: Documentation, https://github.com/Nkzono99/runops/blob/main/docs/reference.md

{runops-2.3.0 → runops-2.3.1}/SPEC.md

@@ -573,6 +573,9 @@ created ──submit──▶ submitted ──(slurm)──▶ running ──▶
 - 対象は `created`、`completed`、`failed`、`cancelled`。`submitted` / `running` は拒否する
   （Slurm が run directory の絶対 path を持つ）。`archived` は `restore` を使う。
 - 既存 path は上書きしない（`move_noreplace`）。`created` の run は job.sh を作り直す。
+- 複数 run は `run.lock` を 1 回取り、全対象の状態と移動先を検査してから移す。1 件でも拒否されれば
+  どれも動かさない。移動中に filesystem の error が起きたら止まり、移し終えた run id を error に示す
+  （exit code 1）。`archive RUN...` も同じ手順で移す。
 - 手で `mv` しても run は id で見つかるが、`run.label` と `[path].run_dir` が古いまま残る。
 
 ---
@@ -967,4 +970,5 @@ v2 の文書は次の 4 ファイルだけにする。
 | 操作範囲は起動 project（2026-10-21） | AGENTS.md block と skill に「対象は起動した project の run と job id だけ。アカウント全体の squeue / sacct は診断に見るだけ。自分で別 directory に移らず、生の `sbatch` / `scancel` を使わない」を書く。CLI は別 project の path を拒否する | 範囲を言わない「再投入して」で、エージェントが別 project や runops 管理外のジョブまで再投入した。cwd で project を決める既存の model に、path 引数の抜け道（`submit --all /other/runs` など）があった。他 session が修正のために止めた run を見分けられるよう、`cancel --why` を note に残し、`retry` はその理由を示して確認を取る（#112） |
 | 置き換えた系列の整理（2026-10-21） | `archive` が DIR（系列）を受け、`--superseded-by` で置き換え先を manifest に記録する。新しい command と tidy の種別は足さない。skill は新系列を作ったら旧系列の要否を確かめ、不要なら archive して現用系列・archive 先・件数を報告する | 修正版を作った後に旧系列が作業領域に残り、毎回ユーザーが整理を頼んでいた。`parent_run` だけでは比較用の派生と置き換えを区別できないので、置き換えは明示操作で記録する。旧 `archive` は 1 件ずつ検査していて、途中の running run で一部だけ移動していた（#113） |
 | 生成ハーネスの書き方（2026-10-21） | AGENTS.md の Rules を「作業の進め方」5 項目に絞り、各項目を「何をするか」と短い理由で書く。skill は手順に専念し、AGENTS.md の方針（manifest を手で編集しない等）を繰り返さない | #112・#113 で規則が増え、禁止形（never / do not）が並んでいた。現行の Claude の prompting 指針は、しないことより何をするかを、理由と共に書くことを勧め、強い禁止語は過剰反応を招くとしている |
+| 複数 run の mv / archive（2026-10-22） | 1 回の `run.lock` の下で全対象を読み直して検査し、通ってから移す。途中の filesystem error では移し終えた run id を示して止まる | `mv A B C --dest X` は 1 件ずつ検査して移していたので、途中の submitted run で一部だけ移動していた。archive は表示の後で lock を取り直していて、その間に再投入された run で同じことが起き得た |
 | 使い方ガイド | `docs/guide.md` を追加し、文書を 4 ファイルにする（§13.3） | README の依頼例の表と command 一覧の間を埋める、場面ごとの流れ（case と run の分け方、smoke から本計算、状態の決まり方、整理）の置き場所が無かった |

{runops-2.3.0 → runops-2.3.1}/docs/reference.md

@@ -162,7 +162,7 @@ run id は削除後も再利用しません。`--dest` に既存 run の中は
 | `runo list [PATH] [--all] [--since 30d] [--case] [--sweep] [--status] [--tag] [--scratch] [--archived] [--json]` | run 一覧 |
 | `runo note RUN [TEXT]` / `runo pin RUN [--off] [--reason]` | ノート追記・表示 / pin |
 | `runo summarize RUN... [--force]` | adapter の要約を `analysis/summary.json` に書く |
-| `runo mv RUN... [--dest DIR] [--label L] [--why TEXT]` | `runs/` 配下で改名・移動。id、入力、sweep、ノートは保つ。submitted / running と archived は不可 |
+| `runo mv RUN... [--dest DIR] [--label L] [--why TEXT]` | `runs/` 配下で改名・移動。id、入力、sweep、ノートは保つ。submitted / running と archived は不可。全対象を検査してから移し、1 件でも不可なら何も動かさない |
 | `runo archive RUN\|DIR... [--superseded-by RUN\|DIR --why TEXT]` / `runo archive --older-than 90d` / `runo restore RUN` | `runs/_archive/` へ移動 / 戻す。移動前に全対象を検査し、指定した RUN が終了していなければ何も動かさない。DIR は配下の run（系列）で、created / submitted / running と pinned を除いて表示する。`--superseded-by` は置き換え先（その run から `--from` で派生した run、無ければ指定した全 run）を `superseded_by` に記録し、`--why` をノートに残す |
 | `runo delete RUN... [-y]` | created・failed・cancelled・scratch の run を削除 |
 | `runo tidy [--apply] [--only KIND] [--older-than AGE] [--purge-work RUN...]` | 整理候補の表示と実行 |

{runops-2.3.0 → runops-2.3.1}/pyproject.toml

@@ -4,7 +4,7 @@ build-backend = "hatchling.build"
 
 [project]
 name = "runops"
-version = "2.3.0"
+version = "2.3.1"
 description = "Run ledger and Slurm job submission for HPC simulation projects"
 readme = "README.md"
 license = "Apache-2.0"

{runops-2.3.0 → runops-2.3.1}/src/runops/__init__.py

@@ -1,3 +1,3 @@
 """runops: run ledger and Slurm job submission for HPC simulation projects."""
 
-__version__ = "2.3.0"
+__version__ = "2.3.1"

{runops-2.3.0 → runops-2.3.1}/src/runops/cli.py

@@ -796,7 +796,7 @@ def archive(
     """Move terminal runs to runs/_archive/ (same relative path)."""
     from datetime import datetime
 
-    from runops.ops.tidy import ArchivePlan, archive_run, plan_archive
+    from runops.ops.tidy import ArchivePlan, archive_runs, plan_archive
 
     project = _project()
     if superseded_by and not why:
@@ -827,11 +827,11 @@ def archive(
             line += f"  -> {', '.join(plan.superseded_by[m.id])}"
         out.print(line)
     _confirm(f"Archive {len(plan.targets)} runs?", yes)
-    for m in plan.targets:
-        dst = archive_run(
-            project, m, superseded_by=plan.superseded_by.get(m.id), why=why, by=_actor()
-        )
-        out.print(f"archived {m.id} -> {project.rel(dst)}")
+    archived = archive_runs(
+        project, plan.targets, superseded_by=plan.superseded_by, why=why, by=_actor()
+    )
+    for run_id, dst in archived:
+        out.print(f"archived {run_id} -> {project.rel(dst)}")
 
 
 @app.command("mv")
@@ -844,7 +844,7 @@ def mv_cmd(
     why: str = typer.Option("", "--why", help="Reason, recorded in the run's notes."),
 ) -> None:
     """Rename or move runs under runs/, keeping their ids, inputs and history."""
-    from runops.ops.tidy import move_run
+    from runops.ops.tidy import move_runs
 
     if not dest and label is None:
         raise UsageError("give --dest and/or --label")
@@ -852,12 +852,10 @@ def mv_cmd(
     targets = _resolve(project, runs)
     if label is not None and len(targets) > 1:
         raise UsageError("--label applies to one RUN; use --dest to move several")
-    warnings: list[str] = []
-    for m in targets:
-        result = move_run(project, m, dest=dest, label=label, why=why, by=_actor())
-        warnings += result.warnings
-        out.print(f"moved {m.id}: {project.rel(result.src)} -> {project.rel(result.dst)}")
-    _warn(warnings)
+    results = move_runs(project, targets, dest=dest, label=label, why=why, by=_actor())
+    for r in results:
+        out.print(f"moved {r.run_id}: {project.rel(r.src)} -> {project.rel(r.dst)}")
+    _warn([w for r in results for w in r.warnings])
 
 
 @app.command()

{runops-2.3.0 → runops-2.3.1}/src/runops/ops/tidy.py

@@ -10,7 +10,7 @@ from pathlib import Path
 
 from runops.adapters.registry import optional
 from runops.case import slugify
-from runops.errors import RunopsError, UsageError
+from runops.errors import RunError, RunopsError, UsageError
 from runops.manifest import ACTIVE_STATES, MANIFEST_FILE, Manifest, RunState
 from runops.ops.common import inside, resolve_dest, run_sim_context
 from runops.ops.submit import write_job_for
@@ -206,12 +206,37 @@ def archive_run(
     """Move a terminal run to ``runs/_archive/``; record replacements and *why* if given."""
     with project.run_lock():
         manifest.reload()
-        return _archive_locked(project, manifest, superseded_by or [], why, by)
+        dst = _archive_dst(project, manifest)
+        return _archive_locked(project, manifest, dst, superseded_by or [], why, by)
 
 
-def _archive_locked(
-    project: Project, manifest: Manifest, superseded_by: list[str], why: str, by: str
-) -> Path:
+def archive_runs(
+    project: Project,
+    manifests: list[Manifest],
+    *,
+    superseded_by: dict[str, list[str]] | None = None,
+    why: str = "",
+    by: str = "",
+) -> list[tuple[str, Path]]:
+    """Archive several runs: all are checked under one ``run.lock`` before any moves."""
+    replacements = superseded_by or {}
+    done: list[tuple[str, Path]] = []
+    with project.run_lock():
+        for m in manifests:
+            m.reload()
+        dsts = [_archive_dst(project, m) for m in manifests]
+        for m, dst in zip(manifests, dsts, strict=True):
+            try:
+                _archive_locked(project, m, dst, replacements.get(m.id, []), why, by)
+            except (RunopsError, OSError) as exc:
+                moved = ", ".join(run_id for run_id, _ in done)
+                tail = f" (already archived: {moved})" if moved else ""
+                raise RunError(f"{m.id}: {exc}{tail}") from exc
+            done.append((m.id, dst))
+    return done
+
+
+def _archive_dst(project: Project, manifest: Manifest) -> Path:
     status = manifest.status
     if status not in _TERMINAL:
         raise UsageError(f"{manifest.id} is {status.value}; only terminal runs can be archived")
@@ -227,6 +252,19 @@ def _archive_locked(
     dst = archive_root / rel
     if dst.exists():
         raise UsageError(f"{dst} already exists")
+    return dst
+
+
+def _archive_locked(
+    project: Project,
+    manifest: Manifest,
+    dst: Path,
+    superseded_by: list[str],
+    why: str,
+    by: str,
+) -> Path:
+    status = manifest.status
+    src = manifest.run_dir.resolve()
     move_noreplace(src, dst)
     moved = Manifest.load(dst / MANIFEST_FILE)
     moved.run["status_before_archive"] = status.value
@@ -284,60 +322,96 @@ class MoveResult:
 _MOVABLE = (RunState.CREATED, RunState.COMPLETED, RunState.FAILED, RunState.CANCELLED)
 
 
-def move_run(
+@dataclass
+class _MovePlan:
+    manifest: Manifest
+    status: RunState
+    src: Path
+    dst: Path
+    label: str
+
+
+def move_runs(
     project: Project,
-    manifest: Manifest,
+    manifests: list[Manifest],
     *,
     dest: str = "",
     label: str | None = None,
     why: str = "",
     by: str = "",
-) -> MoveResult:
-    """Rename and/or move a run under ``runs/``, keeping its id, inputs and history.
-
-    Submitted/running runs are refused (Slurm holds absolute paths into the run
-    directory); archived runs move with ``restore``. The old path and *why* are
-    appended to the run's notes; a created run gets a fresh ``job.sh``.
+) -> list[MoveResult]:
+    """Rename and/or move runs under ``runs/``, keeping their ids, inputs and history.
+
+    Every run is checked under one ``run.lock`` before any directory moves, so a
+    refused run leaves all of them where they were. Submitted/running runs are refused
+    (Slurm holds absolute paths into the run directory); archived runs move with
+    ``restore``. The old path and *why* are appended to each run's notes; a created run
+    gets a fresh ``job.sh``.
     """
+    results: list[MoveResult] = []
     with project.run_lock():
-        manifest.reload()
-        status = manifest.status
-        if status not in _MOVABLE:
-            hint = "restore it first" if status == RunState.ARCHIVED else "wait or cancel it"
-            raise UsageError(f"{manifest.id} is {status.value}; {hint}")
-        src = manifest.run_dir.resolve()
-        parent = resolve_dest(project, dest) if dest else src.parent
-        if inside(parent, src):
-            raise UsageError(f"--dest must not be inside {manifest.id} itself")
-        new_label = (
-            manifest.label
-            if label is None
-            else slugify(label, project.config.defaults.label_max_length)
-        )
-        if label and not new_label:
-            raise UsageError(f"--label {label!r} has no usable characters")
-        name = f"{manifest.id}--{new_label}" if new_label else manifest.id
-        dst = parent / name
-        if dst.resolve() == src:
-            raise UsageError(f"{manifest.id} is already at {project.rel(src)}")
-        if dst.exists():
-            raise UsageError(f"{project.rel(dst)} already exists")
-        parent.mkdir(parents=True, exist_ok=True)
-        move_noreplace(src, dst)
-        moved = Manifest.load(dst / MANIFEST_FILE)
-        moved.run["label"] = new_label
-        moved.section("path")["run_dir"] = project.rel(dst)
-        text = f"moved from {project.rel(src)}"
-        moved.add_note(f"{text}: {why}" if why else text, by or "runops")
-        moved.save()
-        project.update_index(moved.id, dst)
-        result = MoveResult(moved.id, src, dst)
-        if status == RunState.CREATED:
+        plans = [_plan_move(project, m, dest, label) for m in manifests]
+        taken: dict[Path, str] = {}
+        for plan in plans:
+            other = taken.setdefault(plan.dst.resolve(), plan.manifest.id)
+            if other != plan.manifest.id:
+                raise UsageError(
+                    f"{other} and {plan.manifest.id} would both move to {project.rel(plan.dst)}"
+                )
+        for plan in plans:
             try:
-                write_job_for(project, moved)
-            except RunopsError as exc:  # submit regenerates job.sh anyway
-                result.warnings.append(f"{moved.id}: job.sh not regenerated ({exc})")
-    _prune_empty(src.parent, project.runs_dir)
+                results.append(_apply_move(project, plan, why, by))
+            except (RunopsError, OSError) as exc:
+                done = ", ".join(r.run_id for r in results)
+                moved = f" (already moved: {done})" if done else ""
+                raise RunError(f"{plan.manifest.id}: {exc}{moved}") from exc
+            finally:
+                _prune_empty(plan.src.parent, project.runs_dir)
+    return results
+
+
+def _plan_move(project: Project, manifest: Manifest, dest: str, label: str | None) -> _MovePlan:
+    manifest.reload()
+    status = manifest.status
+    if status not in _MOVABLE:
+        hint = "restore it first" if status == RunState.ARCHIVED else "wait or cancel it"
+        raise UsageError(f"{manifest.id} is {status.value}; {hint}")
+    src = manifest.run_dir.resolve()
+    parent = resolve_dest(project, dest) if dest else src.parent
+    if inside(parent, src):
+        raise UsageError(f"--dest must not be inside {manifest.id} itself")
+    new_label = (
+        manifest.label
+        if label is None
+        else slugify(label, project.config.defaults.label_max_length)
+    )
+    if label and not new_label:
+        raise UsageError(f"--label {label!r} has no usable characters")
+    name = f"{manifest.id}--{new_label}" if new_label else manifest.id
+    dst = parent / name
+    if dst.resolve() == src:
+        raise UsageError(f"{manifest.id} is already at {project.rel(src)}")
+    if dst.exists():
+        raise UsageError(f"{project.rel(dst)} already exists")
+    return _MovePlan(manifest, status, src, dst, new_label)
+
+
+def _apply_move(project: Project, plan: _MovePlan, why: str, by: str) -> MoveResult:
+    plan.dst.parent.mkdir(parents=True, exist_ok=True)
+    move_noreplace(plan.src, plan.dst)
+    moved = Manifest.load(plan.dst / MANIFEST_FILE)
+    moved.run["label"] = plan.label
+    moved.section("path")["run_dir"] = project.rel(plan.dst)
+    text = f"moved from {project.rel(plan.src)}"
+    moved.add_note(f"{text}: {why}" if why else text, by or "runops")
+    moved.save()
+    project.update_index(moved.id, plan.dst)
+    result = MoveResult(moved.id, plan.src, plan.dst)
+    if plan.status == RunState.CREATED:
+        try:
+            write_job_for(project, moved)
+        except RunopsError as exc:  # submit regenerates job.sh anyway
+            result.warnings.append(f"{moved.id}: job.sh not regenerated ({exc})")
     return result
 
 

{runops-2.3.0 → runops-2.3.1}/tests/test_archive_series.py

@@ -5,7 +5,11 @@ from __future__ import annotations
 import json
 from pathlib import Path
 
+import pytest
+
+from runops.errors import UsageError
 from runops.manifest import Manifest
+from runops.ops.tidy import archive_runs
 from runops.project import Project
 
 
@@ -96,3 +100,18 @@ def test_archive_series_skips_active_and_pinned(project_dir: Path, cli, fake_slu
     assert cli("archive", b, "-y").code == 0
     res = cli("archive", "runs/demo/mix", "-y")
     assert res.code == 0 and "nothing to archive" in res.out
+
+
+def test_archive_runs_rechecks_under_the_lock(project_dir: Path, cli, fake_slurm) -> None:  # type: ignore[no-untyped-def]
+    a, b = cli(
+        "sweep", "demo", "--set", "x=1,2", "--label", "s", "--apply", "--submit", "-y"
+    ).run_ids()
+    assert cli("cancel", b, "-y").code == 0
+    _finish_all(cli, fake_slurm)
+    project = Project.open(project_dir)
+    stale = [project.resolve_run(a), project.resolve_run(b)]
+    # Between the preview and the move, the cancelled b is retried and resubmitted.
+    assert cli("retry", b, "--submit", "-y").code == 0
+    with pytest.raises(UsageError, match=f"{b} is submitted"):
+        archive_runs(project, stale)
+    assert _manifest(project_dir, a).status.value == "completed"

{runops-2.3.0 → runops-2.3.1}/tests/test_mv_naming.py

@@ -5,8 +5,11 @@ from __future__ import annotations
 import json
 from pathlib import Path
 
+import pytest
+
 from runops.manifest import Manifest
 from runops.project import Project
+from runops.util import move_noreplace
 
 
 def _manifest(project_dir: Path, run_id: str) -> Manifest:
@@ -120,3 +123,40 @@ def test_elapsed_from_exit_record(project_dir: Path, cli, fake_slurm) -> None:
     )
     cli("sync")
     assert _manifest(project_dir, run_id).job["attempts"][-1]["elapsed"] == "1-01:30:05"
+
+
+def test_mv_checks_every_run_before_moving(project_dir: Path, cli, fake_slurm) -> None:  # type: ignore[no-untyped-def]
+    a, b, c = cli("sweep", "demo", "--set", "x=1,2,3", "--label", "s", "--apply").run_ids()
+    before = {r: _manifest(project_dir, r).run_dir for r in (a, b, c)}
+    assert cli("submit", b).code == 0
+    res = cli("mv", a, b, c, "--dest", "regrouped")
+    assert res.code == 2 and f"{b} is submitted" in res.err
+    assert {r: _manifest(project_dir, r).run_dir for r in (a, b, c)} == before
+    assert not (project_dir / "runs/regrouped").exists()
+
+    (project_dir / "runs/regrouped" / before[c].name).mkdir(parents=True)
+    res = cli("mv", a, c, "--dest", "regrouped")
+    assert res.code == 2 and "already exists" in res.err
+    assert _manifest(project_dir, a).run_dir == before[a]
+
+
+def test_mv_reports_runs_moved_before_a_failure(
+    project_dir: Path,
+    cli,  # type: ignore[no-untyped-def]
+    monkeypatch: pytest.MonkeyPatch,
+) -> None:
+    a, b = cli("sweep", "demo", "--set", "x=1,2", "--label", "s", "--apply").run_ids()
+    calls: list[Path] = []
+
+    def flaky(src: Path, dst: Path) -> None:
+        calls.append(src)
+        if len(calls) == 2:
+            raise OSError("disk quota exceeded")
+        move_noreplace(src, dst)
+
+    monkeypatch.setattr("runops.ops.tidy.move_noreplace", flaky)
+    res = cli("mv", a, b, "--dest", "regrouped")
+    assert res.code == 1
+    assert f"{b}: disk quota exceeded (already moved: {a})" in res.err
+    assert _manifest(project_dir, a).run_dir.parent == project_dir / "runs/regrouped"
+    assert _manifest(project_dir, b).run_dir.parent == project_dir / "runs/demo/s"

{runops-2.3.0 → runops-2.3.1}/uv.lock

@@ -525,7 +525,7 @@ wheels = [
 
 [[package]]
 name = "runops"
-version = "2.3.0"
+version = "2.3.1"
 source = { editable = "." }
 dependencies = [
     { name = "jinja2" },