shopdb-flask/plugins/endpoints/services/reconciler.py

"""Warn-only monitor reconciler.

The reconciler answers one question on every agent check-in: for the monitors a
site policy (or the shipped bundle) attaches to this endpoint, is each one the
endpoint is EXPECTED to be running actually reported as running in the latest
snapshot? Every monitor that is expected-running but not observed-running is
DRIFT, and drift is surfaced to an operator for MANUAL intervention.

This is warn-only. The reconciler NEVER starts, stops, or enqueues anything - it
compares observed state against expected state and records the difference. Whether
to later automate remediation (a keep-alive that starts a process when it drifts)
is a separate, deliberate opt-in the product does not ship switched on.

Two seams are kept deliberately narrow so this stays easy to change:

  monitor_expects_running(monitor)
      The single place that decides "should this monitor be up?". Today every
      attached monitor is treated as expected-running (attachment implies intent
      to run). When an observe-only monitor kind arrives - one that only collects
      general state and must never be flagged as drift - this is the one function
      that branches on it. Nothing else in the reconciler encodes that policy.

  compute_drift(asset_id, reported_services)
      Pure given its inputs: it resolves the attached monitors and diffs them
      against the reported running set. It reads no request state and writes no
      rows, so the ingest path can call it inside its own transaction and tests
      can call it directly.
"""

# The running-state vocabulary a watched process/service can report. Kept in
# lockstep with plugins/endpoints/api/routes.py::_RUNNING_STATES so a service the
# panel paints green is never flagged as drift here. Both read from the same
# agent snapshot; disagreeing would show a running app as out-of-desired-state.
_RUNNING_STATES = ('running', 'started', 'up', 'ok', 'active')


def monitor_expects_running(monitor):
    """Whether an attached monitor is expected to be running on an endpoint.

    The seam for the "should be up" decision. Today attachment implies intent to
    run, so every active monitor returns True. A future observe-only monitor kind
    (general state collection, never keep-alive) branches HERE and returns False
    so it is never reported as drift - no other part of the reconciler needs to
    change.
    """
    return bool(getattr(monitor, 'isactive', True))


def _monitor_identity(monitor):
    """The (kind, osname) a monitor watches, or None when it has no identity.

    Mirrors how the services panel reads a monitor's OS identity: a process
    monitor is keyed by its processname, a service monitor by its servicename.
    """
    kind = (getattr(monitor, 'kind', '') or '').lower() or None
    if kind == 'process':
        osname = getattr(monitor, 'processname', None)
    elif kind == 'service':
        osname = getattr(monitor, 'servicename', None)
    else:
        osname = None
    osname = (osname or '').strip()
    if not kind or not osname:
        return None
    return kind, osname


def _reported_running_keys(reported_services):
    """Set of (kind, lowercased-name) the snapshot reports as RUNNING.

    Tolerant of the two shapes the agent snapshot uses: dict entries with
    name/service and state/status (and an optional kind), or bare strings.
    Only running entries land in the set; a stopped or unknown entry is absent,
    which is what makes an expected-running monitor read as drift.
    """
    running = set()
    for entry in reported_services or ():
        if isinstance(entry, dict):
            name = entry.get('name') or entry.get('service') or ''
            raw = (entry.get('state') or entry.get('status') or '').strip().lower()
            kind = (entry.get('kind') or '').lower() or None
        else:
            name, raw, kind = str(entry), '', None
        name = name.strip().lower()
        if not name or raw not in _RUNNING_STATES:
            continue
        running.add((kind, name))
    return running


def _is_running(kind, osname, running_keys):
    """Whether a monitor's identity appears running in the snapshot.

    A snapshot entry may omit kind (older agents, or a plain string list), so a
    match on name alone under an unknown kind still counts as running - the
    reconciler must not invent drift just because the agent did not label a row.
    """
    needle = osname.strip().lower()
    if (kind, needle) in running_keys:
        return True
    if (None, needle) in running_keys:
        return True
    return False


def compute_drift(asset_id, reported_services):
    """Compute the current drift list for an endpoint. Pure, no writes.

    Resolves the monitors attached to the asset's PC type (shipped defaults plus
    site policy), keeps only those that are expected-running, and returns one
    entry per monitor whose identity is not reported as running. Returns an empty
    list when nothing is attached or nothing has drifted, so the caller can store
    it as "computed, no drift" distinct from a null "never computed".

    Isolated from Flask request state: it reads the monitor catalog and the
    passed-in snapshot only, so the ingest path calls it inside its own
    transaction and tests call it directly.
    """
    try:
        from plugins.endpoints.api.routes import _resolved_monitors_for_asset
    except Exception:
        return []

    monitors = _resolved_monitors_for_asset(asset_id) or []
    if not monitors:
        return []

    running_keys = _reported_running_keys(reported_services)

    drift = []
    seen = set()
    for monitor in monitors:
        if not monitor_expects_running(monitor):
            continue
        identity = _monitor_identity(monitor)
        if identity is None:
            continue
        kind, osname = identity
        key = (kind, osname.lower())
        if key in seen:
            continue
        seen.add(key)
        if _is_running(kind, osname, running_keys):
            continue
        drift.append({
            'monitorid': getattr(monitor, 'endpointmonitorid', None),
            'kind': kind,
            'name': osname,
            'displayname': getattr(monitor, 'displayname', None) or osname,
            'expected': 'running',
            'observed': 'notrunning',
        })
    return drift