summaryrefslogtreecommitdiffhomepage
path: root/tests/test_sftp_big.py
diff options
context:
space:
mode:
Diffstat (limited to 'tests/test_sftp_big.py')
-rw-r--r--tests/test_sftp_big.py146
1 files changed, 64 insertions, 82 deletions
diff --git a/tests/test_sftp_big.py b/tests/test_sftp_big.py
index ef12b05c..c58a7912 100644
--- a/tests/test_sftp_big.py
+++ b/tests/test_sftp_big.py
@@ -32,92 +32,74 @@ import unittest
from paramiko.common import o660
-from .test_sftp import make_loopback_sftp
-
FOLDER = os.environ.get('TEST_FOLDER', 'temp-testing000')
-class BigSFTPTest (unittest.TestCase):
-
- def setUp(self):
- global FOLDER
- self.sftp, _ = make_loopback_sftp()
- # TODO: same TODOs as in test_sftp.py re: not doing this awful crap
- for i in range(1000):
- FOLDER = FOLDER[:-3] + '%03d' % i
- try:
- self.sftp.mkdir(FOLDER)
- break
- except (IOError, OSError):
- pass
-
- def tearDown(self):
- self.sftp.rmdir(FOLDER)
-
- def test_1_lots_of_files(self):
+class TestBigSFTP(object):
+ def test_1_lots_of_files(self, sftp):
"""
create a bunch of files over the same session.
"""
numfiles = 100
try:
for i in range(numfiles):
- with self.sftp.open('%s/file%d.txt' % (FOLDER, i), 'w', 1) as f:
+ with sftp.open('%s/file%d.txt' % (FOLDER, i), 'w', 1) as f:
f.write('this is file #%d.\n' % i)
- self.sftp.chmod('%s/file%d.txt' % (FOLDER, i), o660)
+ sftp.chmod('%s/file%d.txt' % (FOLDER, i), o660)
# now make sure every file is there, by creating a list of filenmes
# and reading them in random order.
numlist = list(range(numfiles))
while len(numlist) > 0:
r = numlist[random.randint(0, len(numlist) - 1)]
- with self.sftp.open('%s/file%d.txt' % (FOLDER, r)) as f:
- self.assertEqual(f.readline(), 'this is file #%d.\n' % r)
+ with sftp.open('%s/file%d.txt' % (FOLDER, r)) as f:
+ assert f.readline() == 'this is file #%d.\n' % r
numlist.remove(r)
finally:
for i in range(numfiles):
try:
- self.sftp.remove('%s/file%d.txt' % (FOLDER, i))
+ sftp.remove('%s/file%d.txt' % (FOLDER, i))
except:
pass
- def test_2_big_file(self):
+ def test_2_big_file(self, sftp):
"""
write a 1MB file with no buffering.
"""
kblob = (1024 * b'x')
start = time.time()
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'w') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'w') as f:
for n in range(1024):
f.write(kblob)
if n % 128 == 0:
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
start = time.time()
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
for n in range(1024):
data = f.read(1024)
- self.assertEqual(data, kblob)
+ assert data == kblob
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_3_big_file_pipelined(self):
+ def test_3_big_file_pipelined(self, sftp):
"""
write a 1MB file, with no linefeeds, using pipelining.
"""
kblob = bytes().join([struct.pack('>H', n) for n in range(512)])
start = time.time()
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -125,12 +107,12 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
start = time.time()
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
file_size = f.stat().st_size
f.prefetch(file_size)
@@ -144,18 +126,18 @@ class BigSFTPTest (unittest.TestCase):
chunk = size - n
data = f.read(chunk)
offset = n % 1024
- self.assertEqual(data, k2blob[offset:offset + chunk])
+ assert data == k2blob[offset:offset + chunk]
n += chunk
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_4_prefetch_seek(self):
+ def test_4_prefetch_seek(self, sftp):
kblob = bytes().join([struct.pack('>H', n) for n in range(512)])
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -163,13 +145,13 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
start = time.time()
k2blob = kblob + kblob
chunk = 793
for i in range(10):
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
file_size = f.stat().st_size
f.prefetch(file_size)
base_offset = (512 * 1024) + 17 * random.randint(1000, 2000)
@@ -181,17 +163,17 @@ class BigSFTPTest (unittest.TestCase):
f.seek(offset)
data = f.read(chunk)
n_offset = offset % 1024
- self.assertEqual(data, k2blob[n_offset:n_offset + chunk])
+ assert data == k2blob[n_offset:n_offset + chunk]
offset += chunk
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_5_readv_seek(self):
+ def test_5_readv_seek(self, sftp):
kblob = bytes().join([struct.pack('>H', n) for n in range(512)])
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -199,13 +181,13 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
start = time.time()
k2blob = kblob + kblob
chunk = 793
for i in range(10):
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
base_offset = (512 * 1024) + 17 * random.randint(1000, 2000)
# make a bunch of offsets and put them in random order
offsets = [base_offset + j * chunk for j in range(100)]
@@ -218,20 +200,20 @@ class BigSFTPTest (unittest.TestCase):
for i in range(len(readv_list)):
offset = readv_list[i][0]
n_offset = offset % 1024
- self.assertEqual(next(ret), k2blob[n_offset:n_offset + chunk])
+ assert next(ret) == k2blob[n_offset:n_offset + chunk]
end = time.time()
sys.stderr.write('%ds ' % round(end - start))
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_6_lots_of_prefetching(self):
+ def test_6_lots_of_prefetching(self, sftp):
"""
prefetch a 1MB file a bunch of times, discarding the file object
without using it, to verify that paramiko doesn't get confused.
"""
kblob = (1024 * b'x')
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'w') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'w') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -239,31 +221,31 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
for i in range(10):
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
file_size = f.stat().st_size
f.prefetch(file_size)
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'r') as f:
file_size = f.stat().st_size
f.prefetch(file_size)
for n in range(1024):
data = f.read(1024)
- self.assertEqual(data, kblob)
+ assert data == kblob
if n % 128 == 0:
sys.stderr.write('.')
sys.stderr.write(' ')
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_7_prefetch_readv(self):
+ def test_7_prefetch_readv(self, sftp):
"""
verify that prefetch and readv don't conflict with each other.
"""
kblob = bytes().join([struct.pack('>H', n) for n in range(512)])
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -271,13 +253,13 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
file_size = f.stat().st_size
f.prefetch(file_size)
data = f.read(1024)
- self.assertEqual(data, kblob)
+ assert data == kblob
chunk_size = 793
base_offset = 512 * 1024
@@ -285,22 +267,22 @@ class BigSFTPTest (unittest.TestCase):
chunks = [(base_offset + (chunk_size * i), chunk_size) for i in range(20)]
for data in f.readv(chunks):
offset = base_offset % 1024
- self.assertEqual(chunk_size, len(data))
- self.assertEqual(k2blob[offset:offset + chunk_size], data)
+ assert chunk_size == len(data)
+ assert k2blob[offset:offset + chunk_size] == data
base_offset += chunk_size
sys.stderr.write(' ')
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_8_large_readv(self):
+ def test_8_large_readv(self, sftp):
"""
verify that a very large readv is broken up correctly and still
returned as a single blob.
"""
kblob = bytes().join([struct.pack('>H', n) for n in range(512)])
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'wb') as f:
f.set_pipelined(True)
for n in range(1024):
f.write(kblob)
@@ -308,53 +290,53 @@ class BigSFTPTest (unittest.TestCase):
sys.stderr.write('.')
sys.stderr.write(' ')
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'rb') as f:
data = list(f.readv([(23 * 1024, 128 * 1024)]))
- self.assertEqual(1, len(data))
+ assert len(data) == 1
data = data[0]
- self.assertEqual(128 * 1024, len(data))
+ assert len(data) == 128 * 1024
sys.stderr.write(' ')
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_9_big_file_big_buffer(self):
+ def test_9_big_file_big_buffer(self, sftp):
"""
write a 1MB file, with no linefeeds, and a big buffer.
"""
mblob = (1024 * 1024 * 'x')
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'w', 128 * 1024) as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'w', 128 * 1024) as f:
f.write(mblob)
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
- def test_A_big_file_renegotiate(self):
+ def test_A_big_file_renegotiate(self, sftp):
"""
write a 1MB file, forcing key renegotiation in the middle.
"""
- t = self.sftp.sock.get_transport()
+ t = sftp.sock.get_transport()
t.packetizer.REKEY_BYTES = 512 * 1024
k32blob = (32 * 1024 * 'x')
try:
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'w', 128 * 1024) as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'w', 128 * 1024) as f:
for i in range(32):
f.write(k32blob)
- self.assertEqual(self.sftp.stat('%s/hongry.txt' % FOLDER).st_size, 1024 * 1024)
- self.assertNotEqual(t.H, t.session_id)
+ assert sftp.stat('%s/hongry.txt' % FOLDER).st_size == 1024 * 1024
+ assert t.H != t.session_id
# try to read it too.
- with self.sftp.open('%s/hongry.txt' % FOLDER, 'r', 128 * 1024) as f:
+ with sftp.open('%s/hongry.txt' % FOLDER, 'r', 128 * 1024) as f:
file_size = f.stat().st_size
f.prefetch(file_size)
total = 0
while total < 1024 * 1024:
total += len(f.read(32 * 1024))
finally:
- self.sftp.remove('%s/hongry.txt' % FOLDER)
+ sftp.remove('%s/hongry.txt' % FOLDER)
t.packetizer.REKEY_BYTES = pow(2, 30)