https://github.com/voldemort/voldemort
Tip revision: 051b16f336194550ae45269065daf8f62cc043e6 authored by Vinoth Chandar on 17 May 2012, 22:03:52 UTC
Updated release notes again
Updated release notes again
Tip revision: 051b16f
release_notes.txt
Release 0.90.2 on 05/17/2012
Changes made since 0.90.1
* BDB version upgrade to 4.1.17
* Support for BDB cache mode to cache only index nodes
* Minimize scan impact on BDB cache
* parameters to control lazy cleaner migration, read buffer configuration
* Native backup support
* JNAUtils implementing mlock() of server heap
* Support to restore from replicas in a specific zone
* Donor based rebalancing bug fixes
* Great improvements to Monitoring
* Detailed 95th/99th percentile metrics
* Scan jobs and individual datacleanup jobs
* BDB Mbean monitoring
* Dynamic bandwidth throttling for Read only pull jobs
Release 0.90.1 on 10/10/2011
Changes made since 0.90
* Updated the documentation for Voldemort shell tool in NOTES
* Added Admin API to perform Bdb data cleanup (repairJob)
and corresponding unit tests
* Fixes in restore from replication, store creation code
* Improved failure detector configuration. ThresholdFailureDetector
is now the default option
* Multiple Fixes to the Voldemort ruby client
* Added additional Jmx metrics to expose Bdb environment statistics,
caching statistics and Voldemort batch operation statistics
* Updated default timeout for restore 'from replica' to 365 days
* New feature in the performance tool: '--use-sample' option enables
'read, write back unmodified' transactions in place of writes in
the workload (allows for testing read-write transactions on stores
with complex schemas)
* Added the ability to dynamically update cluster.xml and
reinitialize the scheduler
Release 0.90 on 7/10/2011
Changes made since 0.81
* All upgrading instructions can be found here - https://github.com/voldemort/voldemort/wiki/Upgrading-from-0.81
* Tooling
- Single consolidated administrative tool - Got rid of the admin client
shell and have a better Voldemort admin tool. More documentation -
https://github.com/voldemort/voldemort/wiki/Voldemort-Admin-tool
* Web manager
- Basic GUI in JRuby ( and Sinatra ), capable of querying for store metadata.
- Read contrib/web-manager/README.md for more details
* Rebalancing
- New better rebalancing support with (a) checkpointing (b) progress bar (c) support for RO store rebalancing
- Documentation - https://github.com/voldemort/voldemort/wiki/Voldemort-Rebalancing
- In the process solves Issue 203, 288, 305, 307, 311
* Client side changes
- Pipeline routed store ( with support for topology awareness ) - Changed the default client side routing
to use a state machine like system. More documentation - https://github.com/voldemort/voldemort/wiki/RoutedStore-redesign
- Lazy store client - Some of our clients are very inactive and don't really do many requests. For such
clients it doesn't make sense to bootstrap the metadata at startup. Hence from now onwards clients instantiated
from SocketStoreClientFactory give a LazyStoreClient which will not bootstrap till the first request is made
- Caching store client factory - We have also checked-in a new store client factory which we use internally
at LinkedIn to cache store clients for stores which we repeatedly.
- Failure detector - We have fixed some bugs in the ThresholdFailureDetector. This is an improvement over the naive
BannageFailureDetector which would aggressively mark nodes down after a single failure. More details -
https://github.com/voldemort/voldemort/wiki/Client-side-failure-detector-implementations
- Issue 219: StoreClient::put returns a Version - You may have to upgrade your clients since now the StoreClient returns
a version to be used in successive requests.
* Read-only store changes
- Admin based store swapper - Till 0.81 we relied on a servlet based fetcher / swapper.
Now we support an admin based store swapper which reports progress.
- Other changes - (a) New directory format (b) Two new data format. Support for iterating over read-only data
(c) Checksum of data in .metadata file (d) Some monitoring changes of RO stores
* Jar upgrades
- Introduction of JNA 3.2.7 - We have introduced JNA 3.2.7 in this release for supporting symbolic links in read-only stores.
More details about the new RO format and changes can be found here - https://github.com/voldemort/voldemort/wiki/Upgrading-from-0.81
- Protocol Buffers 2.3.0 - Upgrading protocol buffers from 2.2.0 to 2.3.0
- Avro 1.4.0 - Upgraded Avro from 1.3.0 to 1.4.0
* Storage engine
- Added support for Krati 0.3.6 as a storage engine ( http://sna-projects.com/krati/ ) -
https://github.com/voldemort/voldemort/tree/release-090/contrib/krati
* Core features
- Hinted handoff - https://github.com/voldemort/voldemort/wiki/Hinted-Handoff
- Experimental support for server side transforms - https://github.com/voldemort/voldemort/wiki/Server-side-transforms-in-Voldemort
- Topology awareness ( i.e. datacenter / rack ) - https://github.com/voldemort/voldemort/wiki/Topology-awareness-capability
- Repair Job ( Job which will delete data if the node is not responsible for it )
* Better monitoring
- Tons of JMX changes ( average bytes, etc ) - More details ( https://github.com/voldemort/voldemort/wiki/JMX-Monitoring )
- Key distribution generator - Ability to estimate skew of your cluster ( ./bin/run-class.sh voldemort.utils.KeyDistributionGenerator )
* Clients
- Python - Updated Python client with support for Binary JSON serialized stores ( ./clients/python/README )
- Ruby - Updated Ruby client ( ./clients/ruby/README.md )
Release 0.81 on 6/15/2010
Changes made since 0.80.2:
* IMPORTANT: we have upgraded the Hadoop Core jar to v0.20.2. Since
this version of Hadoop requires Java 6, in order to retain backwards
compatibility with Java 5, you will need to replace this jar with
the Hadoop 0.18.* core jar.
* Multiple donors for Voldemort Rebalancing: speeds up rebalancing, by
allowing a single stealer node to transfer partitions from multiple
nodes
* Features for EC2 Testing
* Read-only Stores: backwards compatibility with 0.70, bug fix in
Checksum code
* Hadoop InputFormat, Pig LoadFunc in Contrib: ability to perform
Map/Reduce (either directly via Hadoop or using Pig) over data in
Voldemort stores. See:
< contrib/hadoop/test/voldemort/hadoop/VoldemortWordCount.java > for
an example.
* Voldemort Performance Tool: performance measurement tool based on
YCSB (Yahoo Cloud Serving Benchmark) code. Run
< bin/voldemort-performance-tool.sh --help > for more information.
* Reverted default failure detector implementation back to
BannagedPeriodFailureDetector due to potential bugs in the
ThresholdFailureDetector
Release 0.80.2 on 4/27/2010
Changes made since 0.80.1:
* Batched NIO writes (improves performance for AdminClient/Streaming
operations when the NIO connector is enabled)
* VoldemortAdminTool: Added support to specify stores, support to
fetch all keys to a binary file, support to save keys in ASCII
(JSON) format [experimental], capability to add stores. Run
< bin/voldemort-admin-tool.sh --help > for more information.
* Minor bug fixes for Rebalancing
* Issue 240: Voldemort fetcher should use different temp directories for
different stores
* Checksum capability during construction of Voldemort read-only stores
Release 0.80.1 on 3/23/2010
Changes made since 0.80:
* Issue 133: Support for Apache Avro as a serialization format
* Issue 223: Changed the default client to use
TresholdFailureDetector
* Fixed issue 222: Revised KeyedResourcePool.close(K key) to fix
leaking sockets
* Fixed issue 198: Revised ReadRepairer to use a separate copy of the
vector clock, fixing a situation where NoSuchElementException would
be thrown
* Miscellaneous enhancement: support for TCP keep-alives, improved
read only store utilities, command line interface to AdminClient,
improved load testing tools
Release 0.80 on 2/18/2010
Changes made since 0.70.1:
* IMPORTANT: backwards compatibility between the client and server has
changed. A backwards incompatibility in the wire protocol was found
between releases 0.60 to 0.70.1 and releases prior to 0.60. We chose
to make 0.80 compatible with 0.57.1 and earlier versions, while
introducing an incompatibility with 0.60-0.70.1. What this means is
that if you're presently running 0.60 and higher, you would need to
upgrade the Voldemort jar files on *all* servers and clients.
* Upgraded the BDB storage engine to use BerkeleyDB-JE 4.0.92,
retaining ability to use BerkeleyDB-JE 3.3.* if desired. IMPORTANT:
if one switches from BerkeleyDB-JE 3.3.* to 4.0.92 they will be able
to access all of existing data. Once a switch has been made to
4.0.92 the data will not be readable by earlier versions of BDB. If
there's a chance that a roll back to 3.3.* might be needed, the best
course of action will be to make a backup of existing data
prior to upgrading.
Switching between 3.3.* and 4.0.* would also require rebuilding the
class files (e.g., by running "ant clean && ant release" after
replacing the BDB jar files).
* Compression support for read-only stores
* Increased the socket buffer size for transferring read-only
stores from Hadoop for improved performance over high-latency links
* NIO support for the Admin Service, including Streaming
Functionality
* Support for adding stores on the fly via the Admin
Service
* Fixed issue 209: Incorrect object passed to List.contains in
RebalanceUtils.getLatestCluster()
* Fixed issue 211: Unnecessary read repairs during getAll() with more
than one key
* Other enhancements: better CLI for rebalancing, throttling in
Admin Service is now based on all disk activity
Release 0.70.1 on 2/1/2010
Changes made since 0.70:
* Fixed issue 205: if no keys passed to getAll() were in partitions
undergoing rebalancing, proxyGetAll() would be called with an
empty list even if rebalancing wasn't happening
Release 0.70 on 1/27/2010
Changes made since 0.60.1:
* A beta of rebalancing (dynamic cluster expansion) support merged
into the main branch. See the project's wiki for more information:
http://wiki.github.com/voldemort/voldemort/voldemort-rebalancing
* New failure detector merged into the main branch:
http://wiki.github.com/voldemort/voldemort/failure-detection
* Beta mechanism for restoring all of node's data from replicas on
demand. This is an alternative to a more gradual mechanism provided
by read-repair: useful when a machine is down for a prolonged period
and is then re-inserted into the cluster.
Invoked via JMX: the operation is restoreDataFromReplication in the
voldemort.server.VoldemortServer MBean, with a mandatory parameter
(integer >= 1) indicating the number of transfers to do in parallel.
* Simple gossip protocol (for cluster metadata) merged
into the main branch. Disabled by default: use "enable.gossip=true"
to enable, use "gossip.interval.ms" to set an interval at which gossip
occurs (default: 30000 i.e., 30 seconds).
* Fixed issue 190: add a way of aggregating performance data over
all stores
* Fixed issue 181: stack trackes shouldn't be filled for Obsolete
version exception
Release 0.60.1 on 12/18/2009
Minor changes made since 0.60:
* Better logging in the exception thrown if config/.temp and
config/.version are copied
* Bumping up the version to 0.60.1 in order to release updated
archives, fixing an error in the stores.xml for single_node_cluster
sample config
Release 0.60 on 12/15/2009
Changes made since 0.57.1:
* Admin Client/Server API: adds support for streaming-based transfer
of entries between nodes, deleting entries on remote nodes,
remotely deleting and updating metadata
* EC2 testing: a way to periodically run integration and performance
tests which involve Voldemort instances on different machines
* Experimental support for views
* Interpolation search for read-only stores
* Support for large lists and strings in the JSON serializer
* LZF compression support
* Ruby client contributed
* Fixed issue 170: hanging if a port is used by another process
* Fixed issue 122: suspicious integer division in
RequestCounter.getThroughput
* Miscellaneous improvements and bug fixes for read-only stores
* Fixed issue 168: added StorageEngine.keys()
Release 0.57.1 on 11/27/2009
Minor change made since 0.57:
* Modified build.xml to exclude .git directory from release tarballs/zipfiles
Release 0.57 on 11/16/2009
The following changes were made since 0.56:
* Fixed an issue in ReadOnlyEngine's close() method
* Fixed hidden logging in StorageService
* Fix for issue 163 (lock mode during get)
* Exposed bdb environment stats with setFast(false)
Release 0.56 on 10/26/2009
The following changes were made since 0.55:
* Fix for issue 164: Changed default bdb.max.logfile.size to 60MB
* Make file deletes asynchronous in read only store swap
* Added better debug logs for bdb stats
* Fixed race condition in AbstractSocketPoolTest
* Added improved monitoring for bdb stats
* Added backoff and retry logic in bootstrap code
* Not logging obosleteVersionException(s) or counting it in JMX exception count
* Fixed issue 159
* C++ client building on OS X
* Rely only on the number of versions returned to decide whether to retrieve
the value for put(K,V)
* Implemented issue 152: getVersion() API in Store
Release 0.55 on 10/7/2009
The following changes were made since 0.52 (in summary):
* Add an event throttler
* Added DataCleanupJob
* Protocol buffers at version 2.2.0
* BDB JE upgraded to 3.3.87
* Added data compression support (enable by adding
<compression>
<type>gzip</type>
</compression>
to value-serializer or key-serializer sections in stores.xml)
* Added a resource pool/socket pool implementation (no longer using
commons-pool)
* Added server-side NIO support (enabled by setting
enable.nio.connector=true
in server.properties)
* Improved the efficiency of the protocol buffers network protocol by using
CodedOutputStream/CodedInputStream (zero copy transfers)
* Fix for issue #21: incorrectness in vector clock inconsistency resolver
* Upgrade google-collections
* Support for building read only stores in Hadoop
* Added a C++ client