Refactor market K-line storage with tiered retention and chunked loading.

Store 1m/5m/1h/12h/1d/1w with per-timeframe policies, aggregate 15m and 2h/4h on read, and support left-pan history fetches via before_ms.

Co-authored-by: Cursor <cursoragent@cursor.com>

hub_ohlcv_lib.py

@@ -3,6 +3,7 @@
 from __future__ import annotations
 
 import math
+import os
 import time
 from typing import Any, Callable, Optional
 
@@ -32,7 +33,26 @@ CHART_TIMEFRAME_ORDER = (
 )
 DAILY_PLUS_TIMEFRAMES = frozenset({"1d", "1w"})
 
-# 部分交易所 ccxt 无原生 12h，或原生 K 线间隔异常时从 1h 聚合
+# 入库 / 同步真源（交易所拉取）
+STORED_TIMEFRAMES = frozenset({"1m", "5m", "1h", "12h", "1d", "1w"})
+PERMANENT_STORED_TIMEFRAMES = frozenset({"12h", "1d", "1w"})
+YEAR_ROLLING_STORED = frozenset({"5m", "1h"})
+
+# 展示周期 → 本地聚合源（不落库）
+CHART_DISPLAY_AGGREGATE_FROM: dict[str, str] = {
+    "15m": "5m",
+    "2h": "1h",
+    "4h": "1h",
+}
+
+SMALL_DISPLAY_TFS = frozenset({"1m", "5m", "15m"})
+MID_DISPLAY_TFS = frozenset({"1h", "2h", "4h"})
+
+HUB_KLINE_1M_MAX_BARS = max(1000, int(os.getenv("HUB_KLINE_1M_MAX_BARS", "10000")))
+HUB_KLINE_5M_1H_RETENTION_DAYS = max(30, int(os.getenv("HUB_KLINE_5M_1H_RETENTION_DAYS", "365")))
+HUB_KLINE_SEED_BARS = max(100, int(os.getenv("HUB_KLINE_SEED_BARS", "500")))
+
+# 部分交易所 ccxt 无原生 12h，或原生 K 线间隔异常时从 1h 聚合（仅远程拉取 fallback）
 OHLCV_AGGREGATE_FROM: dict[str, str] = {
     "12h": "1h",
 }
@@ -55,9 +75,95 @@ def normalize_chart_timeframe(raw: str | None, default: str = "5m") -> str:
     return tf if tf in CHART_TIMEFRAMES else default
 
 
-def bar_limit_for_timeframe(timeframe: str) -> int:
+def sync_timeframe_for_display(timeframe: str) -> str:
+    """展示周期对应的入库 / 同步周期。"""
     tf = normalize_chart_timeframe(timeframe)
-    return 500 if tf in DAILY_PLUS_TIMEFRAMES else 1000
+    return CHART_DISPLAY_AGGREGATE_FROM.get(tf, tf)
+
+
+def aggregation_source_for_display(timeframe: str) -> str | None:
+    tf = normalize_chart_timeframe(timeframe)
+    return CHART_DISPLAY_AGGREGATE_FROM.get(tf)
+
+
+def aggregate_ratio(display_tf: str, source_tf: str) -> int:
+    d = normalize_chart_timeframe(display_tf)
+    s = normalize_chart_timeframe(source_tf)
+    return max(1, int(TIMEFRAME_MS[d] // TIMEFRAME_MS[s]))
+
+
+def chart_initial_limit(timeframe: str) -> int:
+    tf = normalize_chart_timeframe(timeframe)
+    if tf in SMALL_DISPLAY_TFS:
+        return 300
+    if tf == "1w":
+        return 150
+    return 200
+
+
+def chart_chunk_limit(timeframe: str) -> int:
+    tf = normalize_chart_timeframe(timeframe)
+    if tf in SMALL_DISPLAY_TFS:
+        return 500
+    if tf == "1w":
+        return 150
+    if tf in MID_DISPLAY_TFS:
+        return 300
+    return 200
+
+
+def chart_memory_cap(timeframe: str) -> int:
+    tf = normalize_chart_timeframe(timeframe)
+    if tf in SMALL_DISPLAY_TFS:
+        return 5000
+    if tf == "1w":
+        return 500
+    return 1000
+
+
+def bar_limit_for_timeframe(timeframe: str) -> int:
+    return chart_memory_cap(timeframe)
+
+
+def storage_retention_days(storage_tf: str) -> int | None:
+    """None 表示不按天截断（1m 按根数；12h/1d/1w 永久）。"""
+    tf = normalize_chart_timeframe(storage_tf)
+    if tf in YEAR_ROLLING_STORED:
+        return HUB_KLINE_5M_1H_RETENTION_DAYS
+    return None
+
+
+def history_cutoff_ms_for_storage(storage_tf: str, now_ms: int | None = None) -> int:
+    days = storage_retention_days(storage_tf)
+    if days is None:
+        return 0
+    now = int(now_ms if now_ms is not None else time.time() * 1000)
+    return max(0, now - int(days) * 86400000)
+
+
+def seed_bar_target(storage_tf: str) -> int:
+    tf = normalize_chart_timeframe(storage_tf)
+    if tf == "1m":
+        return HUB_KLINE_1M_MAX_BARS
+    if tf in YEAR_ROLLING_STORED:
+        period = TIMEFRAME_MS[tf]
+        return min(
+            int(86400000 * HUB_KLINE_5M_1H_RETENTION_DAYS / period) + 20,
+            150000,
+        )
+    return HUB_KLINE_SEED_BARS
+
+
+def retention_policy_meta() -> dict[str, Any]:
+    return {
+        "1m": {"mode": "bars", "max_bars": HUB_KLINE_1M_MAX_BARS},
+        "5m": {"mode": "days", "days": HUB_KLINE_5M_1H_RETENTION_DAYS},
+        "1h": {"mode": "days", "days": HUB_KLINE_5M_1H_RETENTION_DAYS},
+        "12h": {"mode": "permanent"},
+        "1d": {"mode": "permanent"},
+        "1w": {"mode": "permanent"},
+        "aggregate_from": dict(CHART_DISPLAY_AGGREGATE_FROM),
+    }
 
 
 def last_closed_bar_open_ms(timeframe: str, now_ms: int | None = None) -> int: