changeset 40031:62160d3077cd

cborutil: change buffering strategy Profiling revealed that we were spending a lot of time on the line that was concatenating the old buffer with the incoming data when attempting to decode long byte strings, such as manifest revisions. Essentially, we were feeding N chunks of size len(X) << len(Y) into decode() and continuously allocating a new, larger buffer to hold the undecoded input. This created substantial memory churn and slowed down execution. Changing the code to aggregate pending chunks in a list until we have enough data to fully decode the next atom makes things much more efficient. I don't have exact data, but I recall the old code spending >1s on manifest fulltexts from the mozilla-unified repo. The new code doesn't significantly appear in profile output. Differential Revision: https://phab.mercurial-scm.org/D4854
author Gregory Szorc <gregory.szorc@gmail.com>
date Wed, 03 Oct 2018 09:43:01 -0700
parents e2697acd9381
children f2dffa1359c6
files mercurial/utils/cborutil.py
diffstat 1 files changed, 29 insertions(+), 7 deletions(-) [+]
line wrap: on
line diff
--- a/mercurial/utils/cborutil.py	Wed Oct 03 10:27:44 2018 -0700
+++ b/mercurial/utils/cborutil.py	Wed Oct 03 09:43:01 2018 -0700
@@ -913,7 +913,8 @@
     """
     def __init__(self):
         self._decoder = sansiodecoder()
-        self._leftover = None
+        self._chunks = []
+        self._wanted = 0
 
     def decode(self, b):
         """Attempt to decode bytes to CBOR values.
@@ -924,19 +925,40 @@
         * Integer number of bytes decoded from the new input.
         * Integer number of bytes wanted to decode the next value.
         """
+        # Our strategy for buffering is to aggregate the incoming chunks in a
+        # list until we've received enough data to decode the next item.
+        # This is slightly more complicated than using an ``io.BytesIO``
+        # or continuously concatenating incoming data. However, because it
+        # isn't constantly reallocating backing memory for a growing buffer,
+        # it prevents excessive memory thrashing and is significantly faster,
+        # especially in cases where the percentage of input chunks that don't
+        # decode into a full item is high.
 
-        if self._leftover:
-            oldlen = len(self._leftover)
-            b = self._leftover + b
-            self._leftover = None
+        if self._chunks:
+            # A previous call said we needed N bytes to decode the next item.
+            # But this call doesn't provide enough data. We buffer the incoming
+            # chunk without attempting to decode.
+            if len(b) < self._wanted:
+                self._chunks.append(b)
+                self._wanted -= len(b)
+                return False, 0, self._wanted
+
+            # Else we may have enough data to decode the next item. Aggregate
+            # old data with new and reset the buffer.
+            newlen = len(b)
+            self._chunks.append(b)
+            b = b''.join(self._chunks)
+            self._chunks = []
+            oldlen = len(b) - newlen
+
         else:
-            b = b
             oldlen = 0
 
         available, readcount, wanted = self._decoder.decode(b)
+        self._wanted = wanted
 
         if readcount < len(b):
-            self._leftover = b[readcount:]
+            self._chunks.append(b[readcount:])
 
         return available, readcount - oldlen, wanted