HappyBase Package#

Warning

gRPC is required for using the Cloud Bigtable API. As of May 2016, grpcio is only supported in Python 2.7, so importing gcloud.bigtable in other versions of Python will fail.

Google Cloud Bigtable HappyBase package.

This package is intended to emulate the HappyBase library using Google Cloud Bigtable as the backing store.

Differences in Public API#

Some concepts from HBase/Thrift do not map directly to the Cloud Bigtable API. As a result, the following instance methods and functions could not be implemented:

Connection.enable_table() - no concept of enabled/disabled
Connection.disable_table() - no concept of enabled/disabled
Connection.is_table_enabled() - no concept of enabled/disabled
Connection.compact_table() - table storage is opaque to user
Table.regions() - tables in Cloud Bigtable do not expose internal storage details
Table.counter_set() - method can’t be atomic, so we disable it
The __version__ value for the HappyBase package is None. However, it’s worth nothing this implementation was based off HappyBase 0.9.

In addition, many of the constants from connection are specific to HBase and are defined as None in our module:

COMPAT_MODES
THRIFT_TRANSPORTS
THRIFT_PROTOCOLS
DEFAULT_HOST
DEFAULT_PORT
DEFAULT_TRANSPORT
DEFAULT_COMPAT
DEFAULT_PROTOCOL

Two of these DEFAULT_HOST and DEFAULT_PORT, are even imported in the main happybase package.

Finally, we do not provide the util module. Though it is public in the HappyBase library, it provides no core functionality.

API Behavior Changes#

Since there is no concept of an enabled / disabled table, calling Connection.delete_table() with disable=True can’t be supported. Using that argument will result in a warning.
The Connection constructor disables the use of several arguments and will print a warning if any of them are passed in as keyword arguments. The arguments are:
- host
- port
- compat
- transport
- protocol

In order to make Connection compatible with Cloud Bigtable, we add a instance keyword argument to allow users to pass in their own Instance (which they can construct beforehand).

For example:

from gcloud.bigtable.client import Client
client = Client(project=PROJECT_ID, admin=True)
instance = client.instance(instance_id, location_id)
instance.reload()

from gcloud.bigtable.happybase import Connection
connection = Connection(instance=instance)

Any uses of the wal (Write Ahead Log) argument will result in a warning as well. This includes uses in:
- Batch
- Batch.put()
- Batch.delete()
- Table.put()
- Table.delete()
- Table.batch() factory
When calling Connection.create_table(), the majority of HBase column family options cannot be used. Among
- max_versions
- compression
- in_memory
- bloom_filter_type
- bloom_filter_vector_size
- bloom_filter_nb_hashes
- block_cache_enabled
- time_to_live
Only max_versions and time_to_live are availabe in Cloud Bigtable (as MaxVersionsGCRule and MaxAgeGCRule).

In addition to using a dictionary for specifying column family options, we also accept instances of GarbageCollectionRule or subclasses.
Table.scan() no longer accepts the following arguments (which will result in a warning):
- batch_size
- scan_batching
- sorted_columns
Using a HBase filter string in Table.scan() is not possible with Cloud Bigtable and will result in a TypeError. However, the method now accepts instances of RowFilter and subclasses.
Batch.delete() (and hence Table.delete()) will fail with a ValueError when either a row or column family delete is attempted with a timestamp. This is because the Cloud Bigtable API uses the DeleteFromFamily and DeleteFromRow mutations for these deletes, and neither of these mutations support a timestamp.