GIA/core/workspace/sampling.py

from __future__ import annotations

from datetime import datetime, timezone

# Reasonable server-side caps per graph to keep payload/UI responsive.
DENSITY_POINT_CAPS = {
    "low": 120,
    "medium": 280,
    "high": 560,
}

MS_MINUTE = 60 * 1000
MS_HOUR = 60 * MS_MINUTE
MS_DAY = 24 * MS_HOUR


def _effective_ts_ms(point: dict) -> int:
    value = int(point.get("ts_ms") or 0)
    if value > 0:
        return value
    dt_value = point.get("computed_at")
    if isinstance(dt_value, datetime):
        if dt_value.tzinfo is None:
            dt_value = dt_value.replace(tzinfo=timezone.utc)
        return int(dt_value.timestamp() * 1000)
    return 0


def _bucket_ms_for_age(age_ms: int) -> int:
    if age_ms <= (1 * MS_DAY):
        return 0
    if age_ms <= (7 * MS_DAY):
        return 15 * MS_MINUTE
    if age_ms <= (30 * MS_DAY):
        return 2 * MS_HOUR
    if age_ms <= (180 * MS_DAY):
        return 12 * MS_HOUR
    return 1 * MS_DAY


def _compress_to_target(points: list[dict], target: int) -> list[dict]:
    if target <= 0 or len(points) <= target:
        return points
    if target <= 2:
        return [points[0], points[-1]]
    stride = max(1, int((len(points) - 2) / (target - 2)))
    output = [points[0]]
    idx = 1
    while idx < (len(points) - 1) and len(output) < (target - 1):
        output.append(points[idx])
        idx += stride
    output.append(points[-1])
    return output


def downsample_points(points: list[dict], density: str = "medium") -> list[dict]:
    """
    Tiered time-range downsampling:
    keep high-resolution recent data, progressively bucket older ranges.
    """
    rows = [dict(row or {}) for row in list(points or [])]
    if len(rows) <= 2:
        return rows
    rows.sort(key=_effective_ts_ms)
    latest_ts = _effective_ts_ms(rows[-1])
    buckets: dict[tuple[int, int], dict] = {}
    passthrough: list[dict] = []
    for row in rows:
        ts_ms = _effective_ts_ms(row)
        if ts_ms <= 0:
            continue
        age_ms = max(0, latest_ts - ts_ms)
        bucket_ms = _bucket_ms_for_age(age_ms)
        if bucket_ms <= 0:
            passthrough.append(
                {
                    "x": datetime.fromtimestamp(ts_ms / 1000, tz=timezone.utc).isoformat(),
                    "y": row.get("y"),
                    "ts_ms": ts_ms,
                }
            )
            continue
        bucket_key = (bucket_ms, int(ts_ms // bucket_ms))
        state = buckets.get(bucket_key)
        y_value = row.get("y")
        if state is None:
            buckets[bucket_key] = {
                "count": 1,
                "sum": float(y_value) if y_value is not None else 0.0,
                "last_ts": ts_ms,
                "last_y": y_value,
                "has_value": y_value is not None,
            }
        else:
            state["count"] += 1
            if y_value is not None:
                state["sum"] += float(y_value)
                state["has_value"] = True
            if ts_ms >= int(state["last_ts"]):
                state["last_ts"] = ts_ms
                state["last_y"] = y_value
    output = list(passthrough)
    for state in buckets.values():
        ts_ms = int(state["last_ts"])
        y_value = state["last_y"]
        if bool(state["has_value"]) and int(state["count"]) > 0:
            y_value = round(float(state["sum"]) / float(state["count"]), 3)
        output.append(
            {
                "x": datetime.fromtimestamp(ts_ms / 1000, tz=timezone.utc).isoformat(),
                "y": y_value,
                "ts_ms": ts_ms,
            }
        )
    output.sort(key=lambda row: int(row.get("ts_ms") or 0))
    if not output:
        return []
    # Preserve first/last and reduce to density cap if still too large.
    cap = int(DENSITY_POINT_CAPS.get(str(density or "").strip().lower(), DENSITY_POINT_CAPS["medium"]))
    compact = _compress_to_target(output, cap)
    return compact


def compact_snapshot_rows(snapshot_rows: list[dict], now_ts_ms: int, cutoff_ts_ms: int) -> set[int]:
    """
    Returns IDs to keep using the same age-bucket policy as graph sampling.
    Old rows below cutoff are dropped; remaining rows keep one representative
    per age bucket (latest in bucket), while preserving newest and oldest.
    """
    rows = [dict(row or {}) for row in list(snapshot_rows or [])]
    if not rows:
        return set()
    enriched = []
    for row in rows:
        ts_ms = int(row.get("source_event_ts") or 0)
        if ts_ms <= 0:
            computed_at = row.get("computed_at")
            if isinstance(computed_at, datetime):
                if computed_at.tzinfo is None:
                    computed_at = computed_at.replace(tzinfo=timezone.utc)
                ts_ms = int(computed_at.timestamp() * 1000)
        if ts_ms <= 0:
            continue
        if cutoff_ts_ms > 0 and ts_ms < int(cutoff_ts_ms):
            continue
        enriched.append((int(row.get("id") or 0), ts_ms))
    if not enriched:
        return set()
    enriched.sort(key=lambda item: item[1])
    keep_ids = {enriched[0][0], enriched[-1][0]}
    bucket_map: dict[tuple[int, int], tuple[int, int]] = {}
    latest_ts = int(now_ts_ms or enriched[-1][1])
    for snapshot_id, ts_ms in enriched:
        age_ms = max(0, latest_ts - ts_ms)
        bucket_ms = _bucket_ms_for_age(age_ms)
        if bucket_ms <= 0:
            keep_ids.add(snapshot_id)
            continue
        key = (bucket_ms, int(ts_ms // bucket_ms))
        current = bucket_map.get(key)
        if current is None or ts_ms >= current[1]:
            bucket_map[key] = (snapshot_id, ts_ms)
    for snapshot_id, _ in bucket_map.values():
        keep_ids.add(snapshot_id)
    return keep_ids