Python (2.7 and 3.5+) client for the Iguazio Data Science Platform (the "platform"). Designed to allow fast access to the data layer and basic access to the control layer. This library can be used in Nuclio functions, Jupyter notebooks, local Python IDEs; anywhere with a Python interpreter and access to a platform.
Simply get with pip (locking to a specific version is recommended, as always):
pip install v3io
With the dataplane client you can manipulate data in the platform's multi-model data layer, including:
- Objects
- Key-values (NoSQL)
- Streams
- Containers
Under the hood, the client connects through the platform's web API (https://www.iguazio.com/docs/latest-release/data-layer/reference/web-apis/) and wraps each low level API with an interface. Calls are blocking, but you can use the batching interface to send multiple requests in parallel for greater performance.
Create a dataplane client, passing in the web API endpoint and your access key:
import v3io.dataplane
v3io_client = v3io.dataplane.Client(endpoint='https://v3io-webapi:8081', access_key='some_access_key')Note: In some environments (like Jupyter notebooks) you do not need to pass the endpoint and access key, as they will automatically be inferred from the environment
The client supports a handful of low level APIs, each receiving different arguments but all returning a Response object. A Response object holds 4 fields:
status_code: The HTTP status code returnedoutput: An object containing the parsed response (each API returns a different object)headersandbody: The raw headers and body of the response
You would normally only access the output field unless an API was called that returns raw data like object.get (in which case body holds the response). Consult the reference for each API call to see how to handle its Response object. In the example below, we perform a simple request to get the containers available in our tenant, print the returned status code and containers:
# list the contents of a container
response = v3io_client.container.list('users', '/')
# print the status code. outputs:
#
# Status code: 200
#
print(f'Status code: {response.status_code}')We can also get help information about the parameters this API call receives:
help(v3io_client.container.list)By default, making a request will raise an exception if any non-200 status code is returned. We can override this default behavior in two ways.
The first is to simply never raise an exception and handle the status manually:
# list the contents of a container and never raise an exception
response = v3io_client.container.list('users', '/', raise_for_status=v3io.dataplane.RaiseForStatus.never)
# do anything we want with the status code
# some_logic(response.status_code)The second is to indicate which status codes are acceptable:
# list the contents of a container and raise if the status code is not 200 or 204
response = v3io_client.container.list('users', '/', raise_for_status=[200, 204])To get the highest possible throughput, we can send many requests towards the data layer and wait for all the responses to arrive (rather than send each request and wait for the response). The SDK supports this through batching. Any API call can be made through the client's built in batch object. The API call receives the exact same arguments it would normally receive (except for raise_for_status), and does not block until the response arrives. To wait for all pending responses, call wait() on the batch object:
# do 16 writes in parallel
for idx in range(16):
# returns immediately
v3io_client.batch.object.put(container='bigdata',
path=f'/object{idx}',
body=f'object-{idx}')
# wait for all writes to complete
v3io_client.batch.wait()The looped object.put interface above will send 16 put object requests to the data layer in parallel. When wait is called, it will block until either all responses arrive (in which case it will return a Responses object, containing the responses of each call) or an error occurs - in which case an exception is thrown. You can pass raise_for_status to wait, and it behaves as explained above.
Note: The
batchobject is stateful, so you can only create one batch at a time. However, you can create multiple parallel batches yourself through the client'screate_batch()interface
Put data in an object, get it back and then delete the object:
# put contents to some object
v3io_client.object.put(container='users',
path='/my-object',
body='hello, there')
# get the object
response = v3io_client.object.get(container='users', path='/my-object')
# print the contents. outputs:
#
# hello, there
#
print(response.body.decode('utf-8'))
# delete the object
v3io_client.object.delete(container='users', path='/my-object')For a detailed example, refer to the v3io-objects notebook.
Create a table, update a record and run a query.
items = {
'bob': {'age': 42, 'feature': 'mustache'},
'linda': {'age': 41, 'feature': 'singing'},
'louise': {'age': 9, 'feature': 'bunny ears'},
}
# add the records to the table
for item_key, item_attributes in items.items():
v3io_client.kv.put(container='users', table_path='/bobs-burgers', key=item_key, attributes=item_attributes)
# adds two fields (height, quip) to the louise record
v3io_client.kv.update(container='users',
table_path='/bobs-burgers',
key='louise',
attributes={
'height': 130,
'quip': 'i can smell fear on you'
})
# get a record by key, specifying specific arguments
response = v3io_client.kv.get(container='users',
table_path='/bobs-burgers',
key='louise',
attribute_names=['__size', 'age', 'quip', 'height'])
# print the item from the response. outputs:
#
# {'__size': 0.0, 'quip': 'i can smell fear on you', 'height': 130.0}
#
print(response.output.item)
# create a query, and use an items cursor to iterate the results
items_cursor = v3io_client.kv.new_cursor(container='users',
table_path='/bobs-burgers',
attribute_names=['age', 'feature'],
filter_expression='age > 15')
# print the output
for item in items_cursor.all():
print(item)For a detailed example, refer to the examples/v3io-kv notebook.
Creates a stream with several partitions, writes records to it, reads the records and deletes the stream:
# create a stream w/8 shards
v3io_client.stream.create(container='users',
stream_path='/my-test-stream',
shard_count=8)
# write 4 records - 3 with explicitly specifying the shard and 1 using hashing
records = [
{'shard_id': 1, 'data': 'first shard record #1'},
{'shard_id': 1, 'data': 'first shard record #2'},
{'shard_id': 2, 'data': 'second shard record #1'},
{'data': 'some shard record #1'}
]
v3io_client.stream.put_records(container='users',
stream_path='/my-test-stream',
records=records)
# seek to the beginning of the shard of #1 so we know where to read from
response = v3io_client.stream.seek(container='users',
stream_path='/my-test-stream',
shard_id=1,
seek_type='EARLIEST')
# get records from the shard (should receive 2)
response = v3io_client.stream.get_records(container='users',
stream_path='/my-test-stream',
shard_id=1,
location=response.output.location)
# print the records. outputs:
#
# first shard record #1
# first shard record #2
#
for record in response.output.records:
print(record.data.decode('utf-8'))
# delete the stream
v3io_client.stream.delete(container='users', stream_path='/my-test-stream')For a detailed example, refer to the v3io-streams notebook.
All synchronous APIs are available as async interfaces through the aio module. The differences between the sync and async API is as follows:
- You must initialize a different client (
v3io.aio.dataplane.Client) fromv3io.aio.dataplane - All interfaces should be called with
await v3io.aio.dataplane.RaiseForStatus.nevershould be used overv3io.dataplane.RaiseForStatus.never(although they are the same)- The batching functionality doesn't exist, as you can achieve the same through standard asyncio practices
Note: For the time being, aiohttp must be provided externally - it is not a v3io-py dependency. This will be fixed in future versions.
import v3io.aio.dataplane
v3io_client = v3io.aio.dataplane.Client(endpoint='https://v3io-webapi:8081', access_key='some_access_key')
# put contents to some object
await v3io_client.object.put(container='users',
path='/my-object',
body='hello, there')
# get the object
response = await v3io_client.object.get(container='users', path='/my-object')
# print the contents. outputs:
#
# hello, there
#
print(response.body.decode('utf-8'))
# delete the object
await v3io_client.object.delete(container='users', path='/my-object')To facilitate unit tests, v3io-py can pass along all requests to the user rather than communicate over the network with the dataplane layer. This is done by passing the Client object a verifier transport. This transport receives a list of callables that will be called in the order of requests the Client generates. The user can then verify the content of the request and either return a response or raise an exception. With this mechanism, users can test snippets of their code for various positive and negative behvior.
In the below example, two requests are sent and verified. Using MagicMock for outputs/responses is recommended:
def test_verifier_transport(self):
container_name = 'some_container'
#
# Register a set of verifiers
#
def _verify_object_get(request):
# verify some stuff from the request
self.assertEqual(request.container, container_name)
self.assertEqual(request.path, os.path.join(os.sep, container_name, 'some/path'))
# return a mocked response
return unittest.mock.MagicMock(status_code=200,
body='some body')
def _verify_kv_get(request):
# verify some stuff from the request
self.assertEqual(request.container, container_name)
self.assertEqual(request.path, os.path.join(os.sep, container_name, 'some/table/path/some_item_key'))
# prepare and output mock
output = unittest.mock.MagicMock(item={
'some_key': 'some_value'
})
# prepare a response mock
return unittest.mock.MagicMock(output=output)
# create a verifier transport. pass it a set of request verifiers
verifier_transport = v3io.dataplane.transport.verifier.Transport(request_verifiers=[
_verify_object_get,
_verify_kv_get,
])
#
# Run some flow. This can be a call to the actual code you want to test. By specfiying all
# the verifying behavior beforehand, the actual code under test remains untouched
#
# create a client with a verifier transport
client = v3io.dataplane.Client(transport_kind=verifier_transport)
# do an object get
response = client.object.get(container='some_container', path='some/path')
# verify that we got some body from the verifier
self.assertEqual(response.body, 'some body')
self.assertEqual(response.status_code, 200)
# do an item get
response = client.kv.get(container=container_name,
table_path='some/table/path',
key='some_item_key')
# verify that we got a proper response
self.assertEqual(response.output.item['some_key'], 'some_value')If the verifier transport is inadequate for your use case, a custom transport can be provided. It must conform to the following minimal interface:
class Transport(v3io.dataplane.transport.abstract.Transport):
def __init__(self):
super().__init__(None, '', 0, None, 'DEBUG')
# receives a request and returns a response
def wait_response(self, request, raise_for_status=None):
pass