@smilintux/skcapstone 0.9.0 → 0.12.5

package/src/skcapstone/scheduler_state.py

@@ -0,0 +1,162 @@
+"""Node-local (never-synced) state for the skscheduler.
+
+This module provides per-host, per-job run state that is intentionally kept
+node-local so it never becomes a Syncthing conflict source.  State is stored
+at ``<root>/scheduler/<hostname>/state.json`` and is never replicated.
+"""
+from __future__ import annotations
+
+import json
+import logging
+import threading
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger("skcapstone.scheduler_state")
+
+
+class SchedulerState:
+    """Per-host job state persisted at ``<root>/scheduler/<hostname>/state.json``.
+
+    State is deliberately node-local: the file lives outside any Syncthing-
+    watched subtree so the scheduler never races with sync.  Each instance
+    reads from disk on construction and writes through on every
+    :meth:`record_run` call — there is no in-process cache staleness issue
+    because schedulers are single-process per host.
+
+    Attributes:
+        state_file: Absolute path to the JSON state file for this host.
+    """
+
+    def __init__(self, root: Path, hostname: str) -> None:
+        """Initialise state for ``hostname`` rooted at ``root``.
+
+        Reads any existing state from disk.  If the file is absent or
+        unreadable the state is treated as empty rather than raising.
+
+        Args:
+            root: Repository (or data) root directory.  The state file will
+                be created at ``root/scheduler/<hostname>/state.json``.
+            hostname: Identifier for this node (typically
+                ``socket.gethostname()``).  Used as the directory name so
+                multiple hosts can share the same ``root`` without collision.
+        """
+        self.state_file: Path = Path(root) / "scheduler" / hostname / "state.json"
+        self._data: dict[str, dict] = {}
+        self._write_lock = threading.Lock()
+        if self.state_file.exists():
+            try:
+                self._data = json.loads(
+                    self.state_file.read_text(encoding="utf-8")
+                )
+            except (OSError, json.JSONDecodeError) as exc:
+                logger.warning(
+                    "Could not read scheduler state from %s: %s", self.state_file, exc
+                )
+                self._data = {}
+
+    # ------------------------------------------------------------------
+    # Public API
+    # ------------------------------------------------------------------
+
+    def get(self, job: str) -> dict:
+        """Return the state record for *job*, or a zeroed default.
+
+        The returned dict always contains at least the keys
+        ``run_count``, ``error_count``, and ``last_run``.  Additional keys
+        (``last_status``, ``last_error``) are present once the job has been
+        recorded at least once.
+
+        Args:
+            job: Unique job identifier string.
+
+        Returns:
+            A copy-on-read dict with the job's state.  Mutating the returned
+            dict does **not** persist anything; call :meth:`record_run` to
+            persist changes.
+        """
+        return self._data.get(
+            job, {"run_count": 0, "error_count": 0, "last_run": None}
+        )
+
+    def last_run(self, job: str) -> Optional[datetime]:
+        """Return the timestamp of the most recent run of *job*, or ``None``.
+
+        The returned :class:`~datetime.datetime` is always timezone-aware
+        (UTC) because :meth:`record_run` stores ISO-8601 strings with a
+        ``+00:00`` offset.
+
+        Args:
+            job: Unique job identifier string.
+
+        Returns:
+            A timezone-aware :class:`~datetime.datetime` if the job has run
+            at least once, otherwise ``None``.
+        """
+        raw: Optional[str] = self.get(job).get("last_run")
+        return datetime.fromisoformat(raw) if raw else None
+
+    def record_run(
+        self,
+        job: str,
+        now: Optional[datetime] = None,
+        ok: bool = True,
+        error: str = "",
+    ) -> None:
+        """Record the result of a job execution and persist to disk.
+
+        Increments either ``run_count`` (on success) or ``error_count`` (on
+        failure) and writes the updated state file atomically via
+        :meth:`_flush`.
+
+        Args:
+            job: Unique job identifier string.
+            now: Timestamp for the run.  Defaults to
+                ``datetime.now(timezone.utc)`` when not provided.
+            ok: ``True`` if the job completed successfully, ``False`` on
+                error.
+            error: Human-readable error message.  Ignored when *ok* is
+                ``True``; stored as ``last_error`` otherwise.
+        """
+        ts: datetime = now or datetime.now(timezone.utc)
+        with self._write_lock:
+            rec: dict = self.get(job)
+            rec["last_run"] = ts.isoformat()
+            rec["last_status"] = "ok" if ok else "error"
+            rec["last_error"] = "" if ok else error
+            rec["run_count"] = rec.get("run_count", 0) + (1 if ok else 0)
+            rec["error_count"] = rec.get("error_count", 0) + (0 if ok else 1)
+            self._data[job] = rec
+            self._flush()
+
+    def all(self) -> dict[str, dict]:
+        """Return a shallow copy of all job state records.
+
+        Useful for introspection and dashboards.  Mutations to the returned
+        dict or its values do not affect persisted state.
+
+        Returns:
+            A dict mapping job identifier → state record for every job that
+            has been recorded at least once.
+        """
+        return dict(self._data)
+
+    # ------------------------------------------------------------------
+    # Private helpers
+    # ------------------------------------------------------------------
+
+    def _flush(self) -> None:
+        """Write the in-memory state to :attr:`state_file`.
+
+        Creates parent directories if they do not exist.  Writes the full
+        state dict as indented JSON followed by a trailing newline so the
+        file is human-readable and POSIX-compliant.
+
+        Raises:
+            OSError: If the file cannot be written (e.g. permission denied).
+        """
+        self.state_file.parent.mkdir(parents=True, exist_ok=True)
+        self.state_file.write_text(
+            json.dumps(self._data, indent=2) + "\n", encoding="utf-8"
+        )

package/src/skcapstone/sdk.py

@@ -0,0 +1,310 @@
+"""skcapstone.sdk — the stable public integration facade for sk* services.
+
+This module is the **only** surface that downstream sk* services
+(skmemory, skcomms/skcomms, skchat, sksecurity, capauth, skvoice, skseed,
+cloud9, skgateway, …) should import.  Everything here is semver-tracked and
+will not break across minor releases; the internal modules it wraps
+(``pubsub``, ``scheduler_jobs``, ``coordination``, ``notifications``,
+``service_health``) are NOT part of the public contract and may change freely.
+
+The intended consumer pattern is *optional-by-presence, default-on*::
+
+    try:
+        from skcapstone import sdk as _sk
+        _HAS = (not os.environ.get("SK_STANDALONE")) and _sk.is_available()
+    except ImportError:
+        _sk, _HAS = None, False
+
+    def alert(topic, payload, level="info"):
+        if _HAS:
+            return _sk.alert(f"myservice.{topic}", payload, level=level,
+                             notify=level in ("warn", "error", "critical"))
+        return _native_alert(topic, payload, level)   # service-native fallback
+
+A service that finds skcapstone installed routes alerts through the shared
+PubSub bus and registers scheduled work with the fleet scheduler; a service
+that does not (or that sets ``SK_STANDALONE=1``) keeps using its own
+mechanisms.  See ``docs/ADR-optional-integration-backbone.md``.
+
+Public API:
+    is_available()      -> bool
+    alert(...)          -> bool
+    register_job(...)   -> str (path)
+    unregister_job(...) -> bool
+    coord_create(...)   -> str (task id)
+    register_service(...) -> str (path)
+
+Topic naming convention: ``<service>.<severity>`` (e.g. ``skmemory.error``,
+``sksecurity.critical``).  Severities: ``info | warn | error | critical``.
+"""
+
+from __future__ import annotations
+
+import json
+import logging
+import os
+import uuid
+from datetime import datetime, timezone
+from pathlib import Path
+from typing import Any, Optional
+
+logger = logging.getLogger("skcapstone.sdk")
+
+__all__ = [
+    "is_available",
+    "alert",
+    "register_job",
+    "unregister_job",
+    "coord_create",
+    "register_service",
+    "SEVERITIES",
+]
+
+#: Recognised alert severities, low → high.
+SEVERITIES = ("info", "warn", "error", "critical")
+
+#: Severities that, by convention, also raise a desktop/Telegram notification
+#: when ``notify`` is left at its default.
+_NOTIFY_SEVERITIES = frozenset({"warn", "error", "critical"})
+
+#: severity → desktop notification urgency
+_URGENCY = {
+    "info": "low",
+    "warn": "normal",
+    "error": "normal",
+    "critical": "critical",
+}
+
+
+def _shared_home() -> Path:
+    """Resolve the shared skcapstone root (~/.skcapstone), honouring env."""
+    from . import shared_home  # local import keeps facade import cheap
+
+    return shared_home()
+
+
+def _agent_name() -> str:
+    """Best-effort active agent name, or 'anonymous'."""
+    from . import active_agent_name
+
+    return active_agent_name() or "anonymous"
+
+
+def is_available(require_daemon: bool = False) -> bool:
+    """Return whether skcapstone integration is usable from this process.
+
+    Because the alert bus, scheduler drop-ins and coordination board are all
+    file-based, in-process integration does *not* require the daemon to be
+    running — it only requires that the shared home is resolvable and
+    writable.  ``is_available()`` therefore returns ``True`` whenever the
+    package imported and the home directory can be created.
+
+    Args:
+        require_daemon: When ``True``, additionally probe the local daemon's
+            ``/health`` endpoint and only return ``True`` if it answers.  Use
+            this for capabilities that genuinely need the live daemon (most
+            consumers do not).
+
+    Returns:
+        ``True`` if skcapstone integration can be used, else ``False``.
+    """
+    try:
+        home = _shared_home()
+        home.mkdir(parents=True, exist_ok=True)
+    except Exception as exc:  # pragma: no cover - defensive
+        logger.debug("skcapstone unavailable: %s", exc)
+        return False
+
+    if not require_daemon:
+        return True
+
+    return _daemon_healthy()
+
+
+def _daemon_healthy() -> bool:
+    """Probe the local skcapstone daemon ``/health`` endpoint (best-effort)."""
+    import urllib.request
+
+    port = int(os.environ.get("SKCAPSTONE_PORT", "9383"))
+    url = os.environ.get("SKCAPSTONE_DAEMON_URL", f"http://127.0.0.1:{port}") + "/health"
+    try:
+        with urllib.request.urlopen(url, timeout=1.5) as resp:  # noqa: S310 (localhost)
+            return 200 <= resp.status < 300
+    except Exception as exc:
+        logger.debug("daemon health probe failed (%s): %s", url, exc)
+        return False
+
+
+def alert(
+    topic: str,
+    payload: dict[str, Any],
+    *,
+    level: str = "info",
+    notify: Optional[bool] = None,
+    ttl_seconds: int = 86400,
+) -> bool:
+    """Publish an alert to the shared bus, optionally raising a notification.
+
+    The alert is published to the PubSub topic ``topic`` (callers should use
+    the ``<service>.<severity>`` convention).  When ``notify`` is true — or is
+    left ``None`` and ``level`` is warn/error/critical — a desktop/Telegram
+    notification is also dispatched via the notification manager.
+
+    Args:
+        topic: Fully-qualified topic, e.g. ``"skmemory.error"``.
+        payload: JSON-serialisable event body.
+        level: One of :data:`SEVERITIES`.  Unknown values are treated as
+            ``"info"``.
+        notify: Force notification on/off.  ``None`` (default) means "notify
+            iff severity is warn or higher".
+        ttl_seconds: Message TTL on the bus (default 24h).
+
+    Returns:
+        ``True`` if the message was published (notification is best-effort and
+        does not affect the return value).
+    """
+    if level not in SEVERITIES:
+        level = "info"
+
+    published = False
+    try:
+        from .pubsub import PubSub
+
+        bus = PubSub(_shared_home(), agent_name=_agent_name())
+        bus.publish(topic, dict(payload), ttl_seconds=ttl_seconds, tags=[level])
+        published = True
+    except Exception as exc:
+        logger.warning("sdk.alert publish failed for %r: %s", topic, exc)
+
+    should_notify = (level in _NOTIFY_SEVERITIES) if notify is None else bool(notify)
+    if should_notify:
+        try:
+            from .notifications import notify as _desktop_notify
+
+            summary = payload.get("message") or payload.get("error") or json.dumps(payload)[:200]
+            _desktop_notify(f"[{level}] {topic}", str(summary), _URGENCY.get(level, "normal"))
+        except Exception as exc:  # pragma: no cover - notification is optional
+            logger.debug("sdk.alert notify failed: %s", exc)
+
+    return published
+
+
+def register_job(spec: dict[str, Any], home: Optional[Path] = None) -> str:
+    """Register a scheduled job with the fleet scheduler (jobs.d drop-in).
+
+    Thin wrapper over :func:`skcapstone.scheduler_jobs.register_job`.  The
+    ``spec`` must include a ``name`` and exactly one of ``schedule`` (cron) or
+    ``every`` (interval, e.g. ``"15m"``).  Re-registering the same ``name`` is
+    idempotent, so calling this on every service start is the intended usage.
+
+    Args:
+        spec: Job definition (see jobs.yaml schema).
+        home: Override skcapstone root (defaults to ~/.skcapstone).
+
+    Returns:
+        Filesystem path to the written drop-in fragment, as a string.
+    """
+    from .scheduler_jobs import register_job as _register_job
+
+    return str(_register_job(spec, home=home))
+
+
+def unregister_job(name: str, home: Optional[Path] = None) -> bool:
+    """Remove a previously registered scheduler drop-in.
+
+    Args:
+        name: The job name used at registration.
+        home: Override skcapstone root (defaults to ~/.skcapstone).
+
+    Returns:
+        ``True`` if a fragment existed and was removed.
+    """
+    from .scheduler_jobs import unregister_job as _unregister_job
+
+    return _unregister_job(name, home=home)
+
+
+def coord_create(
+    title: str,
+    *,
+    description: str = "",
+    priority: str = "medium",
+    tags: Optional[list[str]] = None,
+    created_by: str = "",
+    acceptance_criteria: Optional[list[str]] = None,
+    dependencies: Optional[list[str]] = None,
+) -> str:
+    """Create a task on the shared coordination board.
+
+    Args:
+        title: Task title.
+        description: Longer description.
+        priority: ``critical | high | medium | low``.
+        tags: Optional tag list.
+        created_by: Creator name (defaults to the active agent).
+        acceptance_criteria: Optional acceptance bullet list.
+        dependencies: Optional list of blocking task ids.
+
+    Returns:
+        The new task's id.
+    """
+    from .coordination import Board, Task, TaskPriority
+
+    try:
+        prio = TaskPriority(priority)
+    except ValueError:
+        prio = TaskPriority.MEDIUM
+
+    board = Board(_shared_home())
+    task = Task(
+        title=title,
+        description=description,
+        priority=prio,
+        tags=tags or [],
+        created_by=created_by or _agent_name(),
+        acceptance_criteria=acceptance_criteria or [],
+        dependencies=dependencies or [],
+    )
+    board.create_task(task)
+    return task.id
+
+
+def register_service(
+    name: str,
+    health_url: Optional[str] = None,
+    pid_file: Optional[str] = None,
+    home: Optional[Path] = None,
+) -> str:
+    """Advertise a service to skcapstone's discovery registry.
+
+    Writes ``<home>/registry/<name>.json`` describing how to health-check the
+    service.  ``service_health.check_all_services()`` unions these registry
+    entries with its built-in defaults, so a service that calls this on start
+    becomes discoverable without being hardcoded.  Optional — health checks
+    still work with an empty registry.
+
+    Args:
+        name: Service name (unique key).
+        health_url: Optional HTTP URL whose 2xx response means "up".
+        pid_file: Optional pid-file path used as a liveness fallback.
+        home: Override skcapstone root (defaults to ~/.skcapstone).
+
+    Returns:
+        Path to the written registry entry, as a string.
+    """
+    base = Path(home) if home else _shared_home()
+    registry = base / "registry"
+    registry.mkdir(parents=True, exist_ok=True)
+
+    entry = {
+        "name": name,
+        "health_url": health_url,
+        "pid_file": pid_file,
+        "registered_by": _agent_name(),
+        "registered_at": datetime.now(timezone.utc).isoformat(),
+    }
+    final = registry / f"{name}.json"
+    tmp = registry / f".{name}.json.{uuid.uuid4().hex[:8]}.tmp"
+    tmp.write_text(json.dumps(entry, indent=2) + "\n", encoding="utf-8")
+    tmp.rename(final)
+    return str(final)