"""SLO (Service Level Objective) targets for scheduled jobs.

Each registered handler can declare an ``SLOTarget`` describing the
production expectation it commits to:

  * ``success_rate_floor`` — minimum acceptable ok/total ratio over
    the evaluation window (e.g. 0.99 for "less than 1% failure").
  * ``latency_p95_ms_ceiling`` — p95 duration ceiling. None = unbounded.
  * ``max_consecutive_failures`` — most recent N runs all ``failed``
    is a hard breach regardless of overall rate (catches a freshly
    broken job before it drags down the window average).
  * ``tier`` — operational severity: ``P0`` (page immediately, cc
    management), ``P1`` (notify IT admins, dedup signature), ``P2``
    (notify IT admins, lower urgency).

The tier drives ``failure_notifier`` card emoji + recipient fan-out
(P0 also goes to ``management``), and the Prometheus exposition.

Handlers without an explicit ``slo=`` parameter fall through to
``DEFAULT_SLO`` (P2, success_rate >= 0.80, no latency ceiling, max
3 consecutive failures) so every cron has *some* SLO line — the bar
is just lower until an owner declares stricter targets.
"""
from __future__ import annotations

from dataclasses import dataclass
from typing import Literal

# Operational severity tier — drives failure-notifier card emoji,
# recipient fan-out, and the SLO panel color.
SLOTier = Literal["P0", "P1", "P2"]


@dataclass(frozen=True, slots=True)
class SLOTarget:
    """One handler's SLO commitment.

    Immutable + frozen so it's safe to share across cron ticks.
    """
    tier: SLOTier = "P2"
    success_rate_floor: float = 0.80
    latency_p95_ms_ceiling: int | None = None
    max_consecutive_failures: int = 3

    def __post_init__(self) -> None:
        if not (0.0 <= self.success_rate_floor <= 1.0):
            raise ValueError(
                f"success_rate_floor must be in [0.0, 1.0]; got "
                f"{self.success_rate_floor}"
            )
        if self.max_consecutive_failures < 1:
            raise ValueError(
                f"max_consecutive_failures must be >= 1; got "
                f"{self.max_consecutive_failures}"
            )
        if self.latency_p95_ms_ceiling is not None and self.latency_p95_ms_ceiling <= 0:
            raise ValueError(
                f"latency_p95_ms_ceiling must be > 0 when set; got "
                f"{self.latency_p95_ms_ceiling}"
            )
        if self.tier not in ("P0", "P1", "P2"):
            raise ValueError(f"tier must be P0/P1/P2; got {self.tier!r}")


# Default — every registered cron gets *some* SLO line. Owners can
# tighten by passing an explicit ``slo=SLOTarget(...)`` to the
# decorator.
DEFAULT_SLO = SLOTarget(
    tier="P2",
    success_rate_floor=0.80,
    latency_p95_ms_ceiling=None,
    max_consecutive_failures=3,
)


# Pre-defined SLOs for the cron handlers whose breach materially hurts
# the platform. Imported by the handler module at registration time.

# T+0 data-freshness driver — every 15 min, must succeed nearly always.
# Two consecutive failures = data is now visibly stale to IT, page.
SLO_PROCESS_ADVANCE = SLOTarget(
    tier="P0",
    success_rate_floor=0.99,
    latency_p95_ms_ceiling=120_000,  # 2 minutes
    max_consecutive_failures=2,
)

# Notification retry — the platform's last-mile to humans. Breach =
# users stop seeing alerts.
SLO_NOTIFICATION_OUTBOX_RETRY = SLOTarget(
    tier="P1",
    success_rate_floor=0.95,
    latency_p95_ms_ceiling=60_000,
    max_consecutive_failures=3,
)

# Daily ops digest — non-urgent but if it fails for days admins are
# flying blind.
SLO_DAILY_DIGEST = SLOTarget(
    tier="P1",
    success_rate_floor=0.95,
    latency_p95_ms_ceiling=300_000,  # 5 minutes
    max_consecutive_failures=2,
)

# HRIS overdue check — drives the auto-flip to `left`. Soft tier
# because the gate is already conservative.
SLO_OVERDUE_LEAVERS_CHECK = SLOTarget(
    tier="P1",
    success_rate_floor=0.95,
    latency_p95_ms_ceiling=120_000,
    max_consecutive_failures=3,
)

# Cloud cost facts sync — daily, expensive, often partial. Wider
# tolerance.
SLO_CMDB_COST_FACTS_SYNC = SLOTarget(
    tier="P2",
    success_rate_floor=0.85,
    latency_p95_ms_ceiling=15 * 60_000,  # 15 minutes
    max_consecutive_failures=2,
)

# Leaver escalation tick — hourly, MUST keep firing.
SLO_LEAVER_ESCALATION_TICK = SLOTarget(
    tier="P0",
    success_rate_floor=0.99,
    latency_p95_ms_ceiling=60_000,
    max_consecutive_failures=2,
)


__all__ = [
    "DEFAULT_SLO",
    "SLO_CMDB_COST_FACTS_SYNC",
    "SLO_DAILY_DIGEST",
    "SLO_LEAVER_ESCALATION_TICK",
    "SLO_NOTIFICATION_OUTBOX_RETRY",
    "SLO_OVERDUE_LEAVERS_CHECK",
    "SLO_PROCESS_ADVANCE",
    "SLOTarget",
    "SLOTier",
]