On 11-Nov-2011, at 11:50 PM, kishore g wrote: > - There are some things yet to be resolved like if Application Master dies > then all task launched by the ApplicationMaster dies, which is not > desirable for s4 but might be fine for map reduce. > >
I think... if the AM dies or gets wedged..The ApplicationsManager in RM frees the container of AM... re-negotiates another container to restart AM. In case of S4-AM failure...It is possible to only restart the S4-AM... and connect the running S4 instances back to the new ApplicationMaster... The Yarn design doc for reference https://issues.apache.org/jira/secure/attachment/12486023/MapReduce_NextGen_Architecture.pdf Especially the "YARN Availability" section. ./zahoor
