Server crash with bad descriptor data

[prashastk]

Some bug in my code added descriptor blobs which had gone through the following transformations:

np_blob = a 64 dim np float32 array
blob = np.frombuffer(np_blob.tobytes()).tobytes()

After this blob becomes a 32 dim np.float array.

Issue 1 - AddDescriptor for a 64 dim set worked fine with this blob Issue 2 - A FindDescriptor query with k_neighbors crashes the server: Server log

vdms: src/property.cc:44: PMGD::Property::Property(const PMGD::Property&): Assertion `0' failed.

[luisremis]

I have added an extra test for Issue 1 and the it seems the server will correctly catch when the dimension of the blob mismatches. I was not able to reproduce Issue 1.

Can you provide more info?

This is the test I am using to try to replicate.

def test_addSetAndDescriptorsDimMismatch(self):

    db = self.create_connection()

    all_queries = []

    # Add Set
    set_name = "features_64d_dim_mismatched"
    dims = 64
    descriptor_set = {}
    descriptor_set["name"] = set_name
    descriptor_set["dimensions"] = dims

    query = {}
    query["AddDescriptorSet"] = descriptor_set

    all_queries.append(query)

    response, img_array = db.query(all_queries)
    self.assertEqual(response[0]["AddDescriptorSet"]["status"], 0)

    # Add Descriptors
    all_queries = []
    descriptor_blob = []

    x = np.zeros(dims/2)
    x = x.astype('float32')
    # print type(x[0])
    # print "size: ", len(x.tobytes())/4
    descriptor_blob.append(x.tobytes())

    descriptor = {}
    descriptor["set"] = set_name

    query = {}
    query["AddDescriptor"] = descriptor

    all_queries.append(query)

    response, img_array = db.query(all_queries, [descriptor_blob])
    print(db.get_last_response_str())

    # Check correct error
    self.assertEqual(response[0]["status"], -1)
    self.assertEqual(response[0]["info"], "Blob Dimensions Mismatch")

Also, can you provide more detail on Issue 2? Does this happen after the AddDescriptor in Issue 1 is done?

Thanks!

[prashastk]

This should produce Issue 1:

    x = np.zeros(dims/2)
    x = x.astype('float32').tobytes()
    x = np.frombuffer(x.tobytes()).tobytes()
    descriptor_blob.append(x)

(I'll try to re-produce it tomorrow and update this)

Issue 2 was happening after data with the Issue 1 was added into the server.

[luisremis]

I tried that way as well, and get an error:

x = np.frombuffer(x.tobytes()).tobytes()
    AttributeError: 'str' object has no attribute 'tobytes'

I was not able to reproduce this error, even with a couple of different wrong sizes.

I will add more testcases for this issues in #101 for the future, and close this issues unless/until we find a way to reproduce it.

Thanks!