fixed time issue

app/services/news_providers.py

@@ -1,6 +1,7 @@
 import httpx
 from typing import List, Optional, Dict
 from datetime import datetime, timezone, timedelta
+from zoneinfo import ZoneInfo   # stdlib from Python 3.9+ — no extra install needed
 from abc import ABC, abstractmethod
 from app.models import Article
 import os
@@ -87,12 +88,21 @@ class GNewsProvider(NewsProvider):
             query = self.category_map.get(category, category)
             url = f"{self.base_url}/search"
 
-            # Build a window from midnight UTC today to right now.
-            # This is a strict CALENDAR DAY filter, not a rolling 24-hour window.
-            # A job running at 11:59 PM will still only fetch today's articles,
-            # not anything from yesterday.
-            _now    = datetime.now(timezone.utc)
-            _cutoff = _now.replace(hour=0, minute=0, second=0, microsecond=0)  # 00:00:00 UTC today
+            # Build a window from midnight IST today (converted to UTC) to right now.
+            #
+            # Why IST midnight and not UTC midnight?
+            # IST is UTC+5:30. If we used UTC midnight as the "from" date, GNews
+            # would skip articles published in India between 12:00 AM IST and
+            # 5:30 AM IST — the first 5.5 hours of the Indian day.
+            # By computing IST midnight and converting it to UTC, we tell GNews:
+            # "Give me everything published since the Indian day started".
+            _ist_zone   = ZoneInfo("Asia/Kolkata")
+            _now_ist    = datetime.now(_ist_zone)
+            _cutoff_ist = _now_ist.replace(hour=0, minute=0, second=0, microsecond=0)
+            # Convert IST midnight → UTC so the API gets a valid UTC timestamp.
+            _cutoff_utc = _cutoff_ist.astimezone(timezone.utc)
+            # Current moment in UTC for the "to" bound.
+            _now_utc    = datetime.now(timezone.utc)
 
             params = {
                 'q': query,
@@ -100,8 +110,8 @@ class GNewsProvider(NewsProvider):
                 'country': 'us',
                 'max': min(limit, 10),  # GNews free tier max 10
                 'apikey': self.api_key,
-                'from': _cutoff.strftime('%Y-%m-%dT%H:%M:%SZ'),
-                'to':   _now.strftime('%Y-%m-%dT%H:%M:%SZ'),
+                'from': _cutoff_utc.strftime('%Y-%m-%dT%H:%M:%SZ'),  # IST midnight in UTC
+                'to':   _now_utc.strftime('%Y-%m-%dT%H:%M:%SZ'),
             }
             
             async with httpx.AsyncClient(timeout=10.0) as client:
@@ -195,9 +205,14 @@ class NewsAPIProvider(NewsProvider):
             query = self.category_keywords.get(category, category)
             url = f"{self.base_url}/everything"
 
-            # Ask NewsAPI for articles published since midnight UTC today.
-            # Calendar day window: resets cleanly at 00:00:00 UTC every day.
-            _cutoff = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+            # Ask NewsAPI for articles published since midnight IST today.
+            # We compute IST midnight and convert it to UTC before sending it
+            # to the API, because NewsAPI expects UTC timestamps.
+            # This gives Indian users full coverage from their midnight onwards.
+            _ist_zone   = ZoneInfo("Asia/Kolkata")
+            _now_ist    = datetime.now(_ist_zone)
+            _cutoff_ist = _now_ist.replace(hour=0, minute=0, second=0, microsecond=0)
+            _cutoff_utc = _cutoff_ist.astimezone(timezone.utc)  # Convert to UTC for the API
 
             params = {
                 'q': query,
@@ -205,7 +220,7 @@ class NewsAPIProvider(NewsProvider):
                 'sortBy': 'publishedAt',
                 'pageSize': min(limit, 20),
                 'apiKey': self.api_key,
-                'from': _cutoff.strftime('%Y-%m-%dT%H:%M:%SZ'),
+                'from': _cutoff_utc.strftime('%Y-%m-%dT%H:%M:%SZ'),  # IST midnight in UTC
             }
             
             async with httpx.AsyncClient(timeout=10.0) as client:

app/services/scheduler.py

@@ -13,6 +13,7 @@ import pytz
 from app.services.news_aggregator import NewsAggregator
 from app.services.appwrite_db import get_appwrite_db
 from app.services.cache_service import CacheService
+from app.services.upstash_cache import get_upstash_cache   # Needed to bust stale news_v3 keys
 from app.services.adaptive_scheduler import get_adaptive_scheduler, AdaptiveScheduler
 from app.services.research_aggregator import ResearchAggregator
 from app.config import settings
@@ -301,7 +302,31 @@ async def fetch_single_category_job(category: str):
                 invalid_count, irrelevant_count
             )
 
-            # Step 3: Update Redis article cache so the API serves fresh results.
+            # Step 3a: Bust the Upstash news_v3 cache for this category.
+            #
+            # The news route (/api/news/<category>) caches its response in Upstash
+            # under the key  "news_v3:<category>:page:<N>:l<limit>".
+            # Without this delete, a user hitting the page right after an Appwrite
+            # save would still get the stale 5-minute-old response (which may be
+            # empty), because the cache has not expired yet.
+            #
+            # Fix: the moment we save new articles, we surgically delete page-1
+            # of this category's cache. This forces the very next API call to
+            # bypass the cache and read fresh data from Appwrite.
+            if saved_count > 0:
+                try:
+                    upstash = get_upstash_cache()
+                    # Delete the most-visited page (page 1, default limit 20).
+                    # Other pages will expire naturally on their 5-min TTL.
+                    stale_key = f"news_v3:{category}:page:1:l20"
+                    await upstash.delete(stale_key)
+                    logger.info("[CACHE BUST] Deleted stale key '%s' — fresh articles will appear immediately.", stale_key)
+                except Exception as bust_err:
+                    # Cache bust failure is not fatal — articles are already in Appwrite.
+                    # The stale cache will expire on its own in at most 5 minutes.
+                    logger.debug("[CACHE BUST] Could not delete stale key: %s", bust_err)
+
+            # Step 3b: Also update the legacy Redis L1 article cache.
             try:
                 await cache_service.set(f"news:{category}", articles, ttl=settings.CACHE_TTL)
             except Exception as cache_err:

app/utils/data_validation.py

@@ -9,6 +9,7 @@ EMERGENCY HOTFIX (2026-01-23): Fixed AttributeError 'Article' object has no attr
 
 from typing import Dict, Optional, List, Union
 from datetime import datetime, timezone, timedelta
+from zoneinfo import ZoneInfo   # stdlib from Python 3.9+ — no extra install needed
 import re
 from urllib.parse import urlparse
 from dateutil import parser as dateutil_parser
@@ -71,16 +72,21 @@ def is_valid_article(article: Union[Dict, 'Article']) -> bool:
         return False
 
     # ── FRESHNESS GATE ────────────────────────────────────────────────────────
-    # We only want articles published within the last 24 hours.
+    # We only want articles published today, where "today" is measured in
+    # Indian Standard Time (IST = UTC+5:30) — because that is where our
+    # users are.
+    #
+    # Why IST and not UTC?
+    # With UTC midnight as the cutoff, articles published in India between
+    # 12:00 AM IST and 5:30 AM IST (the first 5.5 hours of the Indian day)
+    # were incorrectly rejected, because UTC midnight had not yet arrived.
+    # Switching to IST midnight gives Indian users a full 24-hour day.
     #
     # CRITICAL ORDER: This check runs on the RAW date string, before
     # normalize_article_date() gets a chance to run. That function has a
     # silent fallback: if a date is unparseable it stamps the article with
     # 'right now'. Without this guard, a 3-day-old article with a broken
     # date string would survive normalization and appear fresh.
-    #
-    # Here we do the opposite: if we cannot confidently parse the date,
-    # we reject the article. No date = no entry.
     try:
         if isinstance(raw_date, datetime):
             pub_dt = raw_date
@@ -91,15 +97,18 @@ def is_valid_article(article: Union[Dict, 'Article']) -> bool:
         if pub_dt.tzinfo is None:
             pub_dt = pub_dt.replace(tzinfo=timezone.utc)
 
-        # Midnight UTC today — strict calendar day, not a rolling 24h window.
-        # Example: at 11:59 PM, this is still 00:00:00 of the current day,
-        # so only today's articles pass. Yesterday's articles are rejected
-        # regardless of what time the job fires.
-        cutoff = datetime.now(timezone.utc).replace(hour=0, minute=0, second=0, microsecond=0)
+        # Step 1: Find midnight IST today.
+        # We get the current moment in IST, then zero out hours/minutes/seconds.
+        # This gives us "12:00:00 AM of today in India".
+        ist_zone   = ZoneInfo("Asia/Kolkata")
+        now_ist    = datetime.now(ist_zone)
+        cutoff_ist = now_ist.replace(hour=0, minute=0, second=0, microsecond=0)
 
-        if pub_dt < cutoff:
-            # Article is older than 24 hours — reject it here before any
-            # keyword matching or Redis dedup call wastes time on it.
+        # Step 2: The article timestamp may be in any timezone (UTC, EST, etc.).
+        # Python's datetime comparison handles mixed timezones correctly as long
+        # as both sides are timezone-aware — which they both are here.
+        if pub_dt < cutoff_ist:
+            # Article was published before midnight IST today — reject it.
             return False
 
     except Exception: