Merge pull request #26 from ceph/wip-10066

fixes/improves weird hashing issues on readwrite tests

Reviewed-by: Josh Durgin <jdurgin@redhat.com>

s3tests/readwrite.py

@@ -21,7 +21,7 @@ def reader(bucket, worker_id, file_names, queue, rand):
         objname = rand.choice(file_names)
         key = bucket.new_key(objname)
 
-        fp = realistic.FileVerifier()
+        fp = realistic.FileValidator()
         result = dict(
                 type='r',
                 bucket=bucket.name,
@@ -31,7 +31,7 @@ def reader(bucket, worker_id, file_names, queue, rand):
 
         start = time.time()
         try:
-            key.get_contents_to_file(fp)
+            key.get_contents_to_file(fp._file)
         except gevent.GreenletExit:
             raise
         except Exception as e:
@@ -50,7 +50,7 @@ def reader(bucket, worker_id, file_names, queue, rand):
             end = time.time()
 
             if not fp.valid():
-                m='md5sum check failed start={s} ({se}) end={e} size={sz} obj={o}'.format(s=time.ctime(start), se=start, e=end, sz=fp.size, o=objname)
+                m='md5sum check failed start={s} ({se}) end={e} size={sz} obj={o}'.format(s=time.ctime(start), se=start, e=end, sz=fp._file.tell(), o=objname)
                 result.update(
                     error=dict(
                         msg=m,
@@ -63,13 +63,13 @@ def reader(bucket, worker_id, file_names, queue, rand):
                 result.update(
                     start=start,
                     duration=int(round(elapsed * NANOSECOND)),
-                    chunks=fp.chunks,
                     )
         queue.put(result)
 
 def writer(bucket, worker_id, file_names, files, queue, rand):
     while True:
         fp = next(files)
+        fp.seek(0)
         objname = rand.choice(file_names)
         key = bucket.new_key(objname)
 
@@ -104,7 +104,6 @@ def writer(bucket, worker_id, file_names, files, queue, rand):
             result.update(
                 start=start,
                 duration=int(round(elapsed * NANOSECOND)),
-                chunks=fp.last_chunks,
                 )
 
         queue.put(result)
@@ -189,7 +188,7 @@ def main():
             )
         q = gevent.queue.Queue()
 
-        
+
         # warmup - get initial set of files uploaded if there are any writers specified
         if config.readwrite.writers > 0:
             print "Uploading initial set of {num} files".format(num=config.readwrite.files.num)

s3tests/realistic.py

@@ -21,6 +21,7 @@ def generate_file_contents(size):
     should remove the last 40 chars from the blob to retrieve the original hash
     and binary so that validity can be proved.
     """
+    size = int(size)
     contents = os.urandom(size)
     content_hash = hashlib.sha1(contents).hexdigest()
     return contents + content_hash
@@ -28,22 +29,29 @@ def generate_file_contents(size):
 
 class FileValidator(object):
 
-    def __init__(self, f):
+    def __init__(self, f=None):
         self._file = tempfile.SpooledTemporaryFile()
-        f.seek(0)
+        self.original_hash = None
-        shutil.copyfileobj(f, self._file)
+        self.new_hash = None
-        self.seek(0)
+        if f:
+            f.seek(0)
+            shutil.copyfileobj(f, self._file)
 
     def valid(self):
         """
         Returns True if this file looks valid. The file is valid if the end
         of the file has the md5 digest for the first part of the file.
         """
-        contents = self._file.read()
         self._file.seek(0)
-        original_hash, binary = contents[-40:], contents[:-40]
+        contents = self._file.read()
-        new_hash = hashlib.sha1(binary).hexdigest()
+        self.original_hash, binary = contents[-40:], contents[:-40]
-        return new_hash == original_hash
+        self.new_hash = hashlib.sha1(binary).hexdigest()
+        if not self.new_hash == self.original_hash:
+            print 'original  hash: ', self.original_hash
+            print 'new hash: ', self.new_hash
+            print 'size: ', self._file.tell()
+            return False
+        return True
 
     # XXX not sure if we need all of these
     def seek(self, offset, whence=os.SEEK_SET):
@@ -55,6 +63,10 @@ class FileValidator(object):
     def read(self, size=-1):
         return self._file.read(size)
 
+    def write(self, data):
+        self._file.write(data)
+        self._file.seek(0)
+
 
 class RandomContentFile(object):
     def __init__(self, size, seed):
@@ -228,6 +240,7 @@ def files(mean, stddev, seed=None):
                 break
         yield RandomContentFile(size=size, seed=rand.getrandbits(32))
 
+
 def files2(mean, stddev, seed=None, numfiles=10):
     """
     Yields file objects with effectively random contents, where the
@@ -238,17 +251,17 @@ def files2(mean, stddev, seed=None, numfiles=10):
     stores `numfiles` files and yields them in a loop.
     """
     # pre-compute all the files (and save with TemporaryFiles)
-    rand_files = files(mean, stddev, seed)
     fs = []
     for _ in xrange(numfiles):
-        f = next(rand_files)
         t = tempfile.SpooledTemporaryFile()
-        shutil.copyfileobj(f, t)
+        t.write(generate_file_contents(random.normalvariate(mean, stddev)))
+        t.seek(0)
         fs.append(t)
 
     while True:
         for f in fs:
-            yield PrecomputedContentFile(f)
+            yield f
+
 
 def names(mean, stddev, charset=None, seed=None):
     """

s3tests/tests/test_realistic.py

@@ -40,13 +40,11 @@ class TestFiles(object):
         t = tempfile.SpooledTemporaryFile()
         shutil.copyfileobj(source, t)
         precomputed = realistic.PrecomputedContentFile(t)
-        assert precomputed.valid()
 
-        #verifier = realistic.FileVerifier()
+        verifier = realistic.FileVerifier()
-        #shutil.copyfileobj(precomputed, verifier)
+        shutil.copyfileobj(precomputed, verifier)
 
-        #assert verifier.valid()
+        assert verifier.valid()
-        #assert 0
 
 
 # new implementation