Apache Slider Release 0.50.2 (incubating)¶
August 2014
This release is built against Apache Hadoop 2.4.1, HBase 0.98.4 and Accumulo 1.6.0
Key changes¶
-
Slider now has an integral Chaos Monkey SLIDER-202. This can be configured to start through options in
internal.json
; it will kill a random container or the AM itself based on configured properties. This is intended for use in testing —though may be used in production if desired, and if the application and YARN cluster configured to tolerate the failures. -
Slider runs on Java 6 SLIDER-322. We plan to keep it Java 6-compatible until Hadoop core upgrades to Java 7.
-
Improved Port allocation and management in component instances. (SLIDER-219, SLIDER-273
Incompatible Changes¶
SLIDER-77: use a window for tracking container failures.¶
Previously a simple threshold, "internal.container.failure.threshold"
set the
limit for the number of container failures tolerated for the life of an application.
This has now been reworked to support
a time-bounded window for failures¶
placement in resources.json
as "yarn.container.failure.threshold"
¶
reset/changing during cluster flex¶
Configurable in a combination of days, hours and minutes.¶
This is a major change —and is necessary to support long-lived applications with a slow failure rate, while still detecting and reacting to the situation where many containers are failing in a short period of time.
Because the property name has changed, any cluster where this had been changed from
the default (which is still five) will not pick up the changes. Please use the
new name and set the value in the global section of resources.json
Other changes¶
The full list of completed issues is visible in the Apache JIRA