AdRoll has kindly offered to host the next San Francisco Meetup.
+ Derek Nelson talk about how HBase is used at AdRoll.
+ Francis Liu on Multi-tenant HBase at Y!
+ Sergey Shelukhin of HortonWorks on Compaction Improvements in HBase
+ Benoît Sigoure of Arista Networks on Fast Server Failover for Hadoop and HBase
When machines fail, due to hardware failure, kernel panics, or other, it takes a while for HDFS and HBase to declare the dead node as such. Various timeouts come into play, and these timeouts can't be as aggressive as we'd like, due to the possible false-positives caused by occasional long GC pauses. Such failures can be painful, especially in latency sensitive environments where the 99th percentile matters a lot. Fast Server Failover allows the network to detect machine failures and quickly tell all the other nodes in the cluster that the machine is down, to make failure recovery almost instantaneous. This feature grew out of an actual pain-point while operating a user-facing production HBase cluster.
Pizza and Beer by Cloudera.
(If you would like to talk contact the organizer...)