All Versions
69
Latest Version
Avg Release Cycle
53 days
Latest Release
-

Changelog History
Page 7

  • v1.0.1 Changes

    Feb 19, 2014

    ๐Ÿ› Bug Fixes

    • Include table indexes in KeyspaceMetadata.export_as_string()
    • ๐Ÿ›  Fix broken token awareness on ByteOrderedPartitioner
    • Always close socket when defuncting error'ed connections to avoid a potential file descriptor leak
    • ๐Ÿ– Handle "custom" types (such as the replaced DateType) correctly
    • With libevreactor, correctly handle EAGAIN/EWOULDBLOCK when the message from Cassandra is a multiple of the read buffer size. Previously, if no more data became available to read on the socket, the message would never be processed, resulting in an OperationTimedOut error.
    • 0๏ธโƒฃ Don't break tracing when a Session's row_factory is not the default namedtuple_factory.
    • ๐Ÿ– Handle data that is already utf8-encoded for UTF8Type values
    • ๐Ÿ›  Fix token-aware routing for tokens that fall before the first node token in the ring and tokens that exactly match a node's token
    • Tolerate null source_elapsed values for Trace events. These may not be set when events complete after the main operation has already completed.

    Other

    • Skip sending OPTIONS message on connection creation if compression is disabled or not available and a CQL version has not been explicitly set
    • โž• Add details about errors and the last queried host to OperationTimedOut
  • v1.0.0 Changes

    Jan 29, 2014

    ๐Ÿ› Bug Fixes

    • โฑ Prevent leak of Scheduler thread (even with proper shutdown)
    • Correctly handle ignored hosts, which are common with the DCAwareRoundRobinPolicy
    • Hold strong reference to prepared statement while executing it to avoid garbage collection
    • โž• Add NullHandler logging handler to the cassandra package to avoid warnings about there being no configured logger
    • ๐Ÿ›  Fix bad handling of nodes that have been removed from the cluster
    • Properly escape string types within cql collections
    • ๐Ÿ– Handle setting the same keyspace twice in a row
    • Avoid race condition during schema agreement checks that could result in schema update queries returning before all nodes had seen the change
    • Preserve millisecond-level precision in datetimes when performing inserts with simple (non-prepared) statements
    • Properly defunct connections when libev reports an error by setting errno instead of simply logging the error
    • ๐Ÿ›  Fix endless hanging of some requests when using the libev reactor
    • Always start a reconnection process when we fail to connect to a newly bootstrapped node
    • Generators map to CQL lists, not key sequences
    • Always defunct connections when an internal operation fails
    • Correctly break from handle_write() if nothing was sent (asyncore reactor only)
    • Avoid potential double-erroring of callbacks when a connection becomes defunct

    ๐Ÿ”‹ Features

    • โž• Add default query timeout to Session
    • โž• Add timeout parameter to Session.execute()
    • โž• Add WhiteListRoundRobinPolicy as a load balancing policy option
    • ๐Ÿ‘Œ Support for consistency level LOCAL_ONE
    • ๐Ÿ‘‰ Make the backoff for fetching traces exponentially increasing and configurable

    Other

    • ๐Ÿ‘ป Raise Exception if TokenAwarePolicy is used against a cluster using the Murmur3Partitioner if the murmur3 C extension has not been compiled
    • โž• Add encoder mapping for OrderedDict
    • ๐Ÿ‘‰ Use timeouts on all control connection queries
    • ๐Ÿ’ป Benchmark improvements, including command line options and eay multithreading support
    • ๐Ÿ”’ Reduced lock contention when using the asyncore reactor
    • Warn when non-datetimes are used for 'timestamp' column values in prepared statements
    • โž• Add requirements.txt and test-requirements.txt
    • โœ… TravisCI integration for running unit tests against Python 2.6, Python 2.7, and PyPy
  • v1.0.0-beta4 Changes

    Sep 24, 2013

    ๐Ÿ”‹ Features

    • ๐Ÿ– Handle new blob syntax in Cassandra 2.0 by accepting bytearray objects for blob values
    • ๐Ÿ”– Add cql_version kwarg to Cluster.init

    ๐Ÿ› Bug Fixes

    • ๐Ÿ›  Fix KeyError when building token map with NetworkTopologyStrategy keyspaces (this prevented a Cluster from successfully connecting at all).
    • 0๏ธโƒฃ Don't lose default consitency level from parent PreparedStatement when creating BoundStatements
  • v1.0.0-beta3 Changes

    Sep 20, 2013

    ๐Ÿ”‹ Features

    • ๐Ÿ‘Œ Support for LZ4 compression (Cassandra 2.0+)
    • Token-aware routing will now utilize all replicas for a query instead of just the first replica

    ๐Ÿ› Bug Fixes

    • ๐Ÿ›  Fix libev include path for CentOS
    • ๐Ÿ›  Fix varint packing of the value 0
    • Correctly pack unicode values
    • Don't attempt to return failed connections to the pool when a final result is set
    • ๐Ÿ›  Fix bad iteration of connection credentials
    • ๐Ÿ‘‰ Use blist's orderedset for set collections and OrderedDict for map collections so that Cassandra's ordering is preserved
    • ๐Ÿ›  Fix connection failure on Windows due to unavailability of inet_pton and inet_ntop. (Note that IPv6 inet_address values are still not supported on Windows.)
    • Boolean constants shouldn't be surrounded by single quotes
    • Avoid a potential loss of precision on float constants due to string formatting
    • Actually utilize non-standard ports set on Cluster objects
    • ๐Ÿ›  Fix export of schema as a set of CQL queries

    Other

    • ๐ŸŽ Use cStringIO for connection buffer for better performance
    • Add repr method for Statement classes
    • Raise InvalidTypeParameterError when parameters of the wrong type are used with statements
    • โœ… Make all tests compatible with Python 2.6
    • โž• Add 1s timeout for opening new connections
  • v1.0.0-beta2 Changes

    Aug 19, 2013

    ๐Ÿ› Bug Fixes

    • ๐Ÿ›  Fix pip packaging
  • v1.0.0-beta Changes

    Aug 16, 2013

    ๐ŸŽ‰ Initial release

  • v1.0.0.b7 Changes

    Nov 12, 2013

    ๐Ÿš€ This release makes many stability improvements, especially around prepared statements and node failure handling. In particular, several cases where a request would never be completed (and as a result, leave the application hanging) have been resolved.

    ๐Ÿ”‹ Features

    • โž• Add timeout kwarg to ResponseFuture.result()
    • Create connection pools to all hosts in parallel when initializing new Sesssions.

    ๐Ÿ› Bug Fixes

    • ๐Ÿ‘ป Properly set exception on ResponseFuture when a query fails against all hosts
    • ๐Ÿ‘Œ Improved cleanup and reconnection efforts when reconnection fails on a node that has recently come up
    • ๐Ÿ‘‰ Use correct consistency level when retrying failed operations against a different host. (An invalid consistency level was being used, causing the retry to fail.)
    • ๐Ÿ‘ Better error messages for failed Session.prepare() opertaions
    • Prepare new statements against all hosts in parallel (formerly sequential)
    • ๐Ÿ›  Fix failure to save the new current keyspace on connections. (This could cause problems for prepared statements and lead to extra operations to continuously re-set the keyspace.)
    • Avoid sharing LoadBalancingPolicies across Cluster instances. (When a second Cluster was connected, it effectively mark nodes down for the first Cluster.)
    • ๐Ÿ‘ Better handling of failures during the re-preparation sequence for unrecognized prepared statements
    • Throttle trashing of underutilized connections to avoid trashing newly created connections
    • ๐Ÿ›  Fix race condition which could result in trashed connections being closed before the last operations had completed
    • Avoid preparing statements on the event loop thread (which could lead to deadlock)
    • Correctly mark up non-contact point nodes discovered by the control connection. (This lead to prepared statements not being prepared against those hosts, generating extra traffic later when the statements were executed and unrecognized.)
    • Correctly handle large messages through libev
    • โž• Add timeout to schema agreement check queries
    • More complete (and less contended) locking around manipulation of the pending message deque for libev connections

    Other

    • Prepare statements in batches of 10. (When many prepared statements are in use, this allows the driver to start utilizing nodes that were restarted more quickly.)
    • ๐Ÿ‘ Better debug logging around connection management
    • Don't retain unreferenced prepared statements in the local cache. (If many different prepared statements were created, this would increase memory usage and greatly increase the amount of time required to begin utilizing a node that was added or marked up.)
  • v1.0.0.b6 Changes

    Oct 22, 2013

    ๐Ÿ› Bug Fixes

    • ๐ŸŒฒ Use lazy string formatting when logging
    • Avoid several deadlock scenarios, especially when nodes go down
    • Avoid trashing newly created connections due to insufficient traffic
    • Gracefully handle un-handled Exceptions when erroring callbacks

    Other

    • ๐Ÿšš Node state listeners (which are called when a node is added, removed, goes down, or comes up) should now be registered through Cluster.register_listener() instead of through a host's HealthMonitor (which has been removed)
  • v1.0.0.b5 Changes

    Oct 10, 2013

    ๐Ÿ”‹ Features

    • ๐Ÿ‘ SSL support

    ๐Ÿ› Bug Fixes

    • ๐Ÿ— Avoid KeyError when building replica map for NetworkTopologyStrategy
    • โ†ช Work around python bug which causes deadlock when a thread imports the utf8 module
    • ๐Ÿ– Handle no blist library, which is not compatible with pypy
    • Avoid deadlock triggered by a keyspace being set on a connection (which may happen automatically for new connections)

    Other

    • Switch packaging from Distribute to setuptools, improved C extension support
    • ๐Ÿš€ Use PEP 386 compliant beta and post-release versions