Merge pull request #3275 from lbryio/search_caching_issues

add caching to "search ahead" code and invalidate short_url cache on every block

lbry/wallet/server/db/elasticsearch/search.py

@@ -1,7 +1,7 @@
 import asyncio
 import struct
 from binascii import unhexlify
-from collections import Counter
+from collections import Counter, deque
 from decimal import Decimal
 from operator import itemgetter
 from typing import Optional, List, Iterable, Union
@@ -42,7 +42,7 @@ class SearchIndex:
         self.index = index_prefix + 'claims'
         self.logger = class_logger(__name__, self.__class__.__name__)
         self.claim_cache = LRUCache(2 ** 15)
-        self.short_id_cache = LRUCache(2 ** 17)  # never invalidated, since short ids are forever
+        self.short_id_cache = LRUCache(2 ** 17)
         self.search_cache = LRUCache(2 ** 17)
         self.resolution_cache = LRUCache(2 ** 17)
         self._elastic_host = elastic_host
@@ -134,6 +134,7 @@ class SearchIndex:
                 self.index, body=self.update_filter_query(Censor.RESOLVE, blocked_channels, True), slices=4)
             await self.sync_client.indices.refresh(self.index)
         self.search_cache.clear()
+        self.short_id_cache.clear()
         self.claim_cache.clear()
         self.resolution_cache.clear()
 
@@ -198,7 +199,7 @@ class SearchIndex:
                 self.claim_cache.set(result['claim_id'], result)
 
     async def full_id_from_short_id(self, name, short_id, channel_id=None):
-        key = (channel_id or '') + name + short_id
+        key = '#'.join((channel_id or '', name, short_id))
         if key not in self.short_id_cache:
             query = {'name': name, 'claim_id': short_id}
             if channel_id:
@@ -233,40 +234,56 @@ class SearchIndex:
     async def search_ahead(self, **kwargs):
         # 'limit_claims_per_channel' case. Fetch 1000 results, reorder, slice, inflate and return
         per_channel_per_page = kwargs.pop('limit_claims_per_channel')
-        limit = kwargs.pop('limit', 10)
+        page_size = kwargs.pop('limit', 10)
         offset = kwargs.pop('offset', 0)
         kwargs['limit'] = 1000
+        cache_item = ResultCacheItem.from_cache(f"ahead{per_channel_per_page}{kwargs}", self.search_cache)
+        if cache_item.result is not None:
+            reordered_hits = cache_item.result
+        else:
+            async with cache_item.lock:
+                if cache_item.result:
+                    reordered_hits = cache_item.result
+                else:
                     query = expand_query(**kwargs)
-        result = (await self.search_client.search(
+                    reordered_hits = await self.__search_ahead(query, page_size, per_channel_per_page)
+                    cache_item.result = reordered_hits
+        return list(await self.get_many(*(claim_id for claim_id, _ in reordered_hits[offset:(offset + page_size)])))
+
+    async def __search_ahead(self, query: dict, page_size: int, per_channel_per_page: int):
+        search_hits = deque((await self.search_client.search(
             query, index=self.index, track_total_hits=False, _source_includes=['_id', 'channel_id']
-        ))['hits']
+        ))['hits']['hits'])
-        to_inflate = []
+        reordered_hits = []
-        channel_counter = Counter()
+        channel_counters = Counter()
-        delayed = []
+        next_page_hits_maybe_check_later = deque()
-        while result['hits'] or delayed:
+        while search_hits or next_page_hits_maybe_check_later:
-            if len(to_inflate) % limit == 0:
+            if reordered_hits and len(reordered_hits) % page_size == 0:
-                channel_counter.clear()
+                channel_counters.clear()
+            elif not reordered_hits:
+                pass
             else:
                 break  # means last page was incomplete and we are left with bad replacements
-            new_delayed = []
+            for _ in range(len(next_page_hits_maybe_check_later)):
-            for id, chann in delayed:
+                claim_id, channel_id = next_page_hits_maybe_check_later.popleft()
-                if channel_counter[chann] < per_channel_per_page:
+                if channel_counters[channel_id] < per_channel_per_page:
-                    to_inflate.append((id, chann))
+                    reordered_hits.append((claim_id, channel_id))
-                    channel_counter[hit_channel_id] += 1
+                    channel_counters[channel_id] += 1
                 else:
-                    new_delayed.append((id, chann))
+                    next_page_hits_maybe_check_later.append((claim_id, channel_id))
-            delayed = new_delayed
+            while search_hits:
-            while result['hits']:
+                hit = search_hits.popleft()
-                hit = result['hits'].pop(0)
                 hit_id, hit_channel_id = hit['_id'], hit['_source']['channel_id']
-                if channel_counter[hit_channel_id] < per_channel_per_page:
+                if hit_channel_id is None:
-                    to_inflate.append((hit_id, hit_channel_id))
+                    reordered_hits.append((hit_id, hit_channel_id))
-                    channel_counter[hit_channel_id] += 1
+                elif channel_counters[hit_channel_id] < per_channel_per_page:
-                    if len(to_inflate) % limit == 0:
+                    reordered_hits.append((hit_id, hit_channel_id))
+                    channel_counters[hit_channel_id] += 1
+                    if len(reordered_hits) % page_size == 0:
                         break
                 else:
-                    delayed.append((hit_id, hit_channel_id))
+                    next_page_hits_maybe_check_later.append((hit_id, hit_channel_id))
-        return list(await self.get_many(*(claim_id for claim_id, _ in to_inflate[offset:(offset + limit)])))
+        return reordered_hits
 
     async def resolve_url(self, raw_url):
         if raw_url not in self.resolution_cache: