Merge pull request #2178 from lbryio/lru-cached-read-blob

add a lru cache for decrypted blobs to minimize redownloading them, add `blob_lru_cache_size` to the config to set the cache size

lbrynet/blob/blob_manager.py

@@ -2,6 +2,7 @@ import os
 import typing
 import asyncio
 import logging
+from lbrynet.utils import LRUCache
 from lbrynet.blob.blob_file import is_valid_blobhash, BlobFile, BlobBuffer, AbstractBlob
 from lbrynet.stream.descriptor import StreamDescriptor
 
@@ -30,6 +31,8 @@ class BlobManager:
             else self._node_data_store.completed_blobs
         self.blobs: typing.Dict[str, AbstractBlob] = {}
         self.config = config
+        self.decrypted_blob_lru_cache = None if not self.config.blob_lru_cache_size else LRUCache(
+            self.config.blob_lru_cache_size)
 
     def _get_blob(self, blob_hash: str, length: typing.Optional[int] = None):
         if self.config.save_blobs:

lbrynet/conf.py

@@ -485,7 +485,10 @@ class Config(CLIConfig):
 
     # blob announcement and download
     save_blobs = Toggle("Save encrypted blob files for hosting, otherwise download blobs to memory only.", True)
-
+    blob_lru_cache_size = Integer(
+        "LRU cache size for decrypted downloaded blobs used to minimize re-downloading the same blobs when "
+        "replying to a range request. Set to 0 to disable.", 32
+    )
     announce_head_and_sd_only = Toggle(
         "Announce only the descriptor and first (rather than all) data blob for a stream to the DHT", True,
         previous_names=['announce_head_blobs_only']

lbrynet/stream/downloader.py

@@ -3,7 +3,7 @@ import typing
 import logging
 import binascii
 from lbrynet.error import DownloadSDTimeout
-from lbrynet.utils import resolve_host
+from lbrynet.utils import resolve_host, lru_cache_concurrent
 from lbrynet.stream.descriptor import StreamDescriptor
 from lbrynet.blob_exchange.downloader import BlobDownloader
 from lbrynet.dht.peer import KademliaPeer
@@ -36,6 +36,16 @@ class StreamDownloader:
         self.time_to_descriptor: typing.Optional[float] = None
         self.time_to_first_bytes: typing.Optional[float] = None
 
+        async def cached_read_blob(blob_info: 'BlobInfo') -> bytes:
+            return await self.read_blob(blob_info, 2)
+
+        if self.blob_manager.decrypted_blob_lru_cache:
+            cached_read_blob = lru_cache_concurrent(override_lru_cache=self.blob_manager.decrypted_blob_lru_cache)(
+                cached_read_blob
+            )
+
+        self.cached_read_blob = cached_read_blob
+
     async def add_fixed_peers(self):
         def _delayed_add_fixed_peers():
             self.added_fixed_peers = True

lbrynet/stream/managed_stream.py

@@ -42,6 +42,9 @@ class ManagedStream:
     STATUS_STOPPED = "stopped"
     STATUS_FINISHED = "finished"
 
+    SAVING_ID = 1
+    STREAMING_ID = 2
+
     __slots__ = [
         'loop',
         'config',
@@ -304,7 +307,10 @@ class ManagedStream:
             raise IndexError(start_blob_num)
         for i, blob_info in enumerate(self.descriptor.blobs[start_blob_num:-1]):
             assert i + start_blob_num == blob_info.blob_num
-            decrypted = await self.downloader.read_blob(blob_info, connection_id)
+            if connection_id == self.STREAMING_ID:
+                decrypted = await self.downloader.cached_read_blob(blob_info)
+            else:
+                decrypted = await self.downloader.read_blob(blob_info, connection_id)
             yield (blob_info, decrypted)
 
     async def stream_file(self, request: Request, node: typing.Optional['Node'] = None) -> StreamResponse:
@@ -354,7 +360,7 @@ class ManagedStream:
         self.started_writing.clear()
         try:
             with open(output_path, 'wb') as file_write_handle:
-                async for blob_info, decrypted in self._aiter_read_stream(connection_id=1):
+                async for blob_info, decrypted in self._aiter_read_stream(connection_id=self.SAVING_ID):
                     log.info("write blob %i/%i", blob_info.blob_num + 1, len(self.descriptor.blobs) - 1)
                     await self.loop.run_in_executor(None, self._write_decrypted_blob, file_write_handle, decrypted)
                     self.written_bytes += len(decrypted)

lbrynet/testcase.py

@@ -62,6 +62,7 @@ class CommandTestCase(IntegrationTestCase):
     LEDGER = lbrynet.wallet
     MANAGER = LbryWalletManager
     VERBOSITY = logging.WARN
+    blob_lru_cache_size = 0
 
     async def asyncSetUp(self):
         await super().asyncSetUp()
@@ -81,6 +82,7 @@ class CommandTestCase(IntegrationTestCase):
         conf.lbryum_servers = [('127.0.0.1', 50001)]
         conf.reflector_servers = [('127.0.0.1', 5566)]
         conf.known_dht_nodes = []
+        conf.blob_lru_cache_size = self.blob_lru_cache_size
 
         await self.account.ensure_address_gap()
         address = (await self.account.receiving.get_addresses(limit=1, only_usable=True))[0]

lbrynet/utils.py

@@ -229,11 +229,12 @@ class LRUCache:
         return item in self.cache
 
 
-def lru_cache_concurrent(cache_size: int):
+def lru_cache_concurrent(cache_size: typing.Optional[int] = None,
-    if not cache_size > 0:
+                         override_lru_cache: typing.Optional[LRUCache] = None):
+    if not cache_size and override_lru_cache is None:
         raise ValueError("invalid cache size")
     concurrent_cache = {}
-    lru_cache = LRUCache(cache_size)
+    lru_cache = override_lru_cache or LRUCache(cache_size)
 
     def wrapper(async_fn):
 

tests/integration/test_streaming.py

@@ -2,6 +2,7 @@ import os
 import hashlib
 import aiohttp
 import aiohttp.web
+import asyncio
 
 from lbrynet.utils import aiohttp_request
 from lbrynet.blob.blob_file import MAX_BLOB_SIZE
@@ -373,3 +374,46 @@ class RangeRequests(CommandTestCase):
         await stream.finished_writing.wait()
         with open(stream.full_path, 'rb') as f:
             self.assertEqual(self.data, f.read())
+
+
+class RangeRequestsLRUCache(CommandTestCase):
+    blob_lru_cache_size = 32
+
+    async def _request_stream(self):
+        name = 'foo'
+        url = f'http://{self.daemon.conf.streaming_host}:{self.daemon.conf.streaming_port}/get/{name}'
+
+        async with aiohttp_request('get', url) as req:
+            self.assertEqual(req.headers.get('Content-Type'), 'application/octet-stream')
+            content_range = req.headers.get('Content-Range')
+            content_length = int(req.headers.get('Content-Length'))
+            streamed_bytes = await req.content.read()
+        self.assertEqual(content_length, len(streamed_bytes))
+        self.assertEqual(15, content_length)
+        self.assertEqual(b'hi\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00\x00', streamed_bytes)
+        self.assertEqual('bytes 0-14/15', content_range)
+
+    async def test_range_requests_with_blob_lru_cache(self):
+        self.data = b'hi'
+        self.daemon.conf.save_blobs = False
+        self.daemon.conf.save_files = False
+        await self.stream_create('foo', '0.01', data=self.data, file_size=0)
+        await self.daemon.jsonrpc_file_list()[0].fully_reflected.wait()
+        await self.daemon.jsonrpc_file_delete(delete_from_download_dir=True, claim_name='foo')
+        self.assertEqual(0, len(os.listdir(self.daemon.blob_manager.blob_dir)))
+
+        await self.daemon.streaming_runner.setup()
+        site = aiohttp.web.TCPSite(self.daemon.streaming_runner, self.daemon.conf.streaming_host,
+                                   self.daemon.conf.streaming_port)
+        await site.start()
+        self.assertListEqual(self.daemon.jsonrpc_file_list(), [])
+
+        await self._request_stream()
+        self.assertEqual(1, len(self.daemon.jsonrpc_file_list()))
+        self.server.stop_server()
+
+        # running with cache size 0 gets through without errors without
+        # this since the server doesnt stop immediately
+        await asyncio.sleep(1, loop=self.loop)
+
+        await self._request_stream()