ivakegg commented on a change in pull request #78: update 1.9.0 release notes URL: https://github.com/apache/accumulo-website/pull/78#discussion_r182468631
########## File path: _posts/release/2018-03-06-accumulo-1.9.0.md ########## @@ -22,15 +22,47 @@ Below are resources for this release: In [ACCUMULO-4611], public API in ClientConfiguration using commons config types was deprecated to better support Hadoop 3 in the future. New methods were added to replace these methods which cause -this release to be a 1.9.0 release. +this release to be a 1.9.0 release. These changes allow removal of commons config from Accumulo's +API in 2.0.0. If using ClientConfiguration, then switching from existing constructors to the new +static methods [create()], [fromFile()], or [fromMap()] will ensure your code works in 2.0.0. ### Performance Improvements -Several peformance improvements were added in [#379], [ACCUMULO-4778], [ACCUMULO-4779] and -[ACCUMULO-4781]. +Accumulo was profiled while running lots of concurrent small scans. During this excercise these +performance bugs were found and fixed: [#378], [ACCUMULO-4778], [ACCUMULO-4779], [ACCUMULO-4781], +[ACCUMULO-4782], [ACCUMULO-4788], [ACCUMULO-4789], [ACCUMULO-4790], [ACCUMULO-4797], +[ACCUMULO-4798], [ACCUMULO-4799], [ACCUMULO-4800], [ACCUMULO-4801], [ACCUMULO-4805], and [ACCUMULO-4809] + +### Fixed upgrade process to set version on all volumes + +During upgrades, only one volume in a multiple HDFS volume was updated with the correct version. This would cause all tablet servers +to complain and ultimately fail. [ACCUMULO-4686] fixes this by setting the version on all volumes. + +### Updated Accumulo to work with new releases of Guava + +In [ACCUMULO-4702], dependencies on Beta-annotated Guava classes and methods were removed. While Accumulo still +includes Guava 14 in its tarball, it will work with newer versions of Guava in client code. It has been tested to work +with Guava 23. + +### Updated RFile to prevent very large blocks + +RFiles now use windowed statistics ([ACCUMULO-4669]) to prevent very large blocks. In 1.8.0 a bug was introduced that +caused RFile data block sizes to grow very large in the case where key sizes slowly increased. This could lead +to degraded query performance or out of memory exceptions on tablet servers. ## Other Notable Changes +* [#403] - Enabled more metrics reporting +* [#410] - Fixed inefficient auths check +* [ACCUMULO-4528] - Add import/export table info to docs +* [ACCUMULO-4636] - System iterator perfomance improvements +* [ACCUMULO-4643] - Allow iterators to yield Review comment: Not too sure how much information is needed, but here is a shot at some text for ACCUMULO-4643: We have added the capability for an iterator implementation to yield control back to the tserver in a seek or next call prior to finding the first or next top key/value. This is to be used to avoid starvation of other scans when iterators are being used that may take a very large amount of time to find keys to return. This was done by providing a YieldingKeyValueIterator interface to be implemented on top of the SortedKeyValueIterator interface. A callback mechanism is used to provide the last key looked at when yielding which will be supplied on a subsequent seek call. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected] With regards, Apache Git Services
