Uploaded byHBaseCon
HBaseCon 2015: DeathStar - Easy, Dynamic, Multi-tenant HBase via YARN
The document discusses the implementation of Deathstar, a dynamic, multi-tenant HBase framework running on YARN, addressing challenges such as capacity planning, application customization, and resource utilization. It highlights a provisioning model that supports both dynamic and static clusters, enabling easier HBase upgrades, scaling, and isolation within a common HDFS layer. Key challenges like logging and scheduling constraints are acknowledged, with ongoing debugging and potential solutions proposed.
More Related Content
Similar to HBaseCon 2015: DeathStar - Easy, Dynamic, Multi-tenant HBase via YARN
![Discover.hdp2.2.h base.final[2]](https://cdn.slidesharecdn.com/ss_thumbnails/discover-141218155001-conversion-gate02-thumbnail.jpg?width=640&height=640&fit=bounds)
HBaseCon 2015: DeathStar - Easy, Dynamic, Multi-tenant HBase via YARN
- 1. DeathStar Easy, Dynamic, Multi-tenantHBase via YARN Ishan Chhabra, Nitin Aggarwal Rocketfuel Inc.
- 2. In a notso distant past…
- 3.
- 4.
- 5.
- 6. Hard to capacityplan or support new applications
- 7. The Common Solution: SeparateClusters
- 8. Non uniform networkusage Different DFSs’, leading to lot of copying of data Low cluster utilization High lead time for new applications
- 9. Run HBase onYARN Built on top of Slider Solution: DeathStar
- 10. Hangar App Cluster 3 Provisioning Model App Cluster2 App Cluster 1
- 11. (grid/deathstar): $ gitcommit Capacity planning and configuration discussion Create simple JSON config As applications mature from hangar to their cluster
- 12. Dynamic Cluster: Make APIcall to start, stop and scale cluster Static Cluster: Good to go
- 14.
- 15.
- 16.
- 17.
- 18. Clusters out ofthin air
- 19. Hot swap anew cluster (human error / corruption) Easier HBase version upgrades and testing Temporary scale up for backfill “Dynamic” enables interesting use cases
- 20. Key Challenges andSolutions
- 21. Another failure mode Takencare of by auto restarts RM HA in the works
- 22. Early Days: Bugs Slider didnot acknowledge container allocations correctly Fixed scheduled for 0.8: SLIDER-828
- 23. Not easily reproducible, stilldebugging Zombie Regionservers Early Days: Bugs
- 24. Long running appsa secondary use case Logging, an unsolved problem Store logs on local disks, considering ELK
- 25. YARN/Slider lack certain schedulingconstraints At most x instances per node for spread and availability Custom patch in-house
- 26. Rolling restarts for configchanges Unsolved. SLIDER-226
- 27.
- 28. Metrics Reporting Custom hadoop metrics OpenTSDBreporter App name passed via config
- 29.
- 30.
- 31. Thank you! Questions? Reach usat: ishan@rocketfuel.com naggarwal@rocketfuel.com
- 32. Are you sureyou want to scroll down?
- 33.
- 34. Key Insight: HBase Multi-Tenancy andAccess Patterns
- 35.
- 36. HBase Data Pipeline 1 MutableMaterialized View Stream 1 Data Pipeline 2 Data Pipeline 3 Stream 2
- 37. HBase Prep Stage Transient Cache Stage 1Stage 2 Stage 3
