Merge pull request #374 from xylar/updates-for-unified

Refactor mache.deploy for E3SM-Unified deployment support

Author: xylar

docs/users_guide/deploy.md

@@ -199,9 +199,17 @@ Important settings:
 - `machines.path`: optional path to target-owned machine config files.
 - `pixi.prefix`: required in practice.
 - `pixi.channels`: required and must be non-empty.
+- `pixi.login_mpi`: optional alternate MPI setting for login nodes. When set,
+  `mache deploy run` may create a second pixi environment for login-node use
+  and the generated load script will auto-select between the login and compute
+  environments.
+- `pixi.login_prefix`: optional install path for that login-node pixi
+  environment. If omitted, mache reuses `pixi.prefix` when the login and
+  compute MPI settings match, or defaults to `<pixi.prefix>_login` when they
+  differ.
 - `permissions.group`: optional shared-filesystem group to apply after a
   successful deploy. If unset, `mache deploy run` falls back to merged machine
-  config `[deploy] group`, then legacy `[e3sm_unified] group`.
+  config `[deploy] group`.
 - `permissions.world_readable`: optional boolean controlling whether deployed
   files are readable outside the shared group. Hooks may override both values
   through `ctx.runtime["permissions"]`.
@@ -557,12 +565,15 @@ created.
 5. Resolves toolchain pairs for Spack.
 6. Renders `deploy/pixi.toml.j2` into the install prefix and runs
    `pixi install`.
-7. Optionally installs a development copy of `mache` when a fork/branch was
+7. Optionally creates a second login-node pixi environment when
+   `pixi.login_mpi` is configured or a hook sets `ctx.runtime["pixi"]`
+   accordingly.
+8. Optionally installs a development copy of `mache` when a fork/branch was
    requested.
-8. Optionally builds and installs JIGSAW.
-9. Optionally deploys or loads Spack environments.
-10. Optionally installs the target software in editable mode.
-11. Writes one or more `load_<software>*.sh` scripts.
+9. Optionally builds and installs JIGSAW.
+10. Optionally deploys or loads Spack environments.
+11. Optionally installs the target software in editable mode.
+12. Writes one or more `load_<software>*.sh` scripts.
 
 When a toolchain pair is selected, script names include machine, compiler,
 and MPI tags, for example:
@@ -572,6 +583,11 @@ and MPI tags, for example:
 Those load scripts are the main artifact a downstream user consumes after the
 deployment completes.
 
+When a login pixi environment is configured, the generated load script chooses
+the login environment on login nodes and the compute environment inside common
+batch jobs such as Slurm, PBS, or Cobalt. Spack activation remains compute-only
+in that split-environment case.
+
 ## The command-line contract
 
 There are really two CLIs involved:
@@ -621,8 +637,10 @@ Examples from the current contract:
 - `--bootstrap-only` is deploy-only.
 - `--machine` is deploy-plus-run.
 - `--mache-version` is bootstrap-plus-run.
-- `--pixi`, `--prefix`, `--recreate`, `--quiet`, `--mache-fork`, and
+- `--pixi`, `--pixi-path`, `--recreate`, `--quiet`, `--mache-fork`, and
   `--mache-branch` are routed across all relevant phases.
+- `--prefix` is still accepted as a deprecated compatibility alias for
+  `--pixi-path`.
 
 ### How `deploy/custom_cli_spec.json` works
 

mache/deploy/bootstrap.py

@@ -30,6 +30,8 @@
 )
 BOOTSTRAP_SETUPTOOLS_SPEC = '>=60'
 BOOTSTRAP_WHEEL_SPEC = '*'
+CONDA_FORGE_LABEL_ROOT = 'https://conda.anaconda.org/conda-forge/label'
+MACHE_DEV_LABEL = f'{CONDA_FORGE_LABEL_ROOT}/mache_dev'
 
 
 def check_call(
@@ -409,6 +411,8 @@ def _run(log_filename):
     if args.recreate and (bootstrap_dir / '.pixi').exists():
         shutil.rmtree(bootstrap_dir / '.pixi')
 
+    _write_bootstrap_pixi_config(bootstrap_dir=bootstrap_dir)
+
     # Create/update the bootstrap env
     if args.mache_fork is not None and args.mache_branch is not None:
         _clone_mache_repo(
@@ -488,11 +492,13 @@ def _parse_args():
         'on PATH.',
     )
     parser.add_argument(
+        '--pixi-path',
         '--prefix',
-        dest='prefix',
-        help='Install the environment into this prefix (directory). '
+        dest='pixi_path',
+        help='Install the pixi environment at this path (directory). '
         'This is a deploy-time option; bootstrap accepts it for CLI '
-        'contract compatibility but does not use it.',
+        'contract compatibility but does not use it. `--prefix` is '
+        'deprecated.',
     )
     parser.add_argument(
         '--recreate',
@@ -766,10 +772,11 @@ def _write_bootstrap_pixi_toml_with_mache(
     python_version,
 ):
     name = f'{software}-mache-bootstrap'
+    channels = _get_bootstrap_channels_for_mache_version(mache_version)
     lines = [
         '[workspace]',
         f'name = "{name}"',
-        'channels = ["conda-forge"]',
+        f'channels = [{", ".join(json.dumps(c) for c in channels)}]',
         f'platforms = ["{_get_pixi_platform()}"]',
         'channel-priority = "strict"',
         '',
@@ -782,6 +789,30 @@ def _write_bootstrap_pixi_toml_with_mache(
     pixi_toml_path.write_text('\n'.join(lines) + '\n', encoding='utf-8')
 
 
+def _write_bootstrap_pixi_config(*, bootstrap_dir: Path) -> None:
+    config_dir = bootstrap_dir / '.pixi'
+    config_dir.mkdir(parents=True, exist_ok=True)
+    config_toml = config_dir / 'config.toml'
+    lines = [
+        '# Keep conda-forge label channels on Anaconda because prefix.dev',
+        '# does not currently mirror public conda-forge labels such as ',
+        '# mache_dev.',
+        '[mirrors]',
+        f'"{CONDA_FORGE_LABEL_ROOT}" = [',
+        f'  "{CONDA_FORGE_LABEL_ROOT}",',
+        ']',
+    ]
+    config_toml.write_text('\n'.join(lines) + '\n', encoding='utf-8')
+
+
+def _get_bootstrap_channels_for_mache_version(mache_version: str) -> List[str]:
+    normalized = str(mache_version).strip().lower()
+    if 'rc' in normalized:
+        return [MACHE_DEV_LABEL, 'conda-forge']
+
+    return ['conda-forge']
+
+
 def _format_pixi_version_specifier(version: str) -> str:
     """Return a pixi version specifier from a mache version pin.
 

mache/deploy/machine.py

@@ -32,7 +32,7 @@ def get_known_mache_machines(*, package: str = 'mache.machines') -> set[str]:
         if not name.endswith('.cfg'):
             continue
         machine = name[:-4]
-        if machine == 'default':
+        if machine == 'default' or machine.startswith('default-'):
             continue
         known.add(machine)
     return known
@@ -70,7 +70,7 @@ def get_known_target_machines(*, machines_path: str | None) -> set[str]:
         if not entry.endswith('.cfg'):
             continue
         machine = entry[:-4]
-        if machine == 'default':
+        if machine == 'default' or machine.startswith('default-'):
             continue
         known.add(machine)
     return known