Data Infrastructure at LinkedIn
Download as PPTX, PDF•43 likes•11,747 views
The document outlines LinkedIn's data infrastructure, detailing its online, near-line, and offline data processing systems, and addressing the challenges and solutions in managing large datasets. It highlights components like Databus for consistent data changes and Kafka for messaging, while discussing workflow management and data integration issues faced by data scientists. Key takeaways emphasize the importance of balancing specialized and generic solutions for effective infrastructure in a rapidly growing environment.
Data Infrastructure at LinkedIn
- 1. Data Infrastructure at Linkedin Jun Rao and Sam Shah LinkedIn Confidential ©2013 All Rights Reserved
- 2. Outline LinkedIn Confidential ©2013 All Rights Reserved 2 1. LinkedIn introduction 2. Online/nearline infrastructure overview 3. Infrastructure for data mining 4. Conclusion
- 3. The World‟s Largest Professional Network Members Worldwide 2 new Members Per Second 100M+ Monthly Unique Visitors 200M+ 2M+ Company Pages Connecting Talent Opportunity. At scale… LinkedIn Confidential ©2013 All Rights Reserved 3
- 4. 4 Member Profiles Large dataset Medium writes Very high reads Freshness <1s
- 5. People You May Know 5 Large dataset Compute intensive High reads Freshness ~hrs
- 6. LinkedIn Today 6 Moving dataset High writes High reads Freshness ~mins
- 7. The Big-Data Feedback Loop LinkedIn Confidential ©2013 All Rights Reserved 7 Value Insights Scale Product ScienceData Member Engagement Virality Signals Refinement Infrastructure Analytics
- 8. LinkedIn Data Infrastructure: Three-Phase Abstraction LinkedIn Confidential ©2013 All Rights Reserved 8 Users Online Data Infra Near-Line Infra Application Offline Data Infra Infrastructure Latency & Freshness Requirements Products Online Activity that should be reflected immediately • Member Profiles • Company Profiles • Connections • Messages • Endorsements • Skills Near-Line Activity that should be reflected soon • Activity Streams • Profile Standardization • News • Recommendations • Search • Messages Offline Activity that can be reflected later • People You May Know • Connection Strength • News • Recommendations • Next best idea…
- 9. LinkedIn Data Infrastructure: Sample Stack 9 Infra challenges in 3-phase ecosystem are diverse, complex and specific Some off-the-shelf. Significant investment in home-grown, deep and interesting platforms
- 11. Databus : Timeline-Consistent Change Data Capture LinkedIn Data Infrastructure Solutions
- 12. Databus at LinkedIn 12 DB Bootstrap Capture Changes On-line Changes On-line Changes DB Consistent Snapshot at U Transport independent of data source: Oracle, MySQL, … Transactional semantics In order, at least once delivery Tens of relays Hundreds of sources Low latency - milliseconds Consumer 1 Consumer n Client Databus ClientLib Consumer 1 Consumer n Databus ClientLib Client Relay Event Win
- 14. Voldemort: Highly-Available Distributed KV Store LinkedIn Data Infrastructure Solutions 14
- 15. • Pluggable components • Tunable consistency / availability • Key/value model, server side “views” • 10 clusters, 100+ nodes • Largest cluster – 10K+ qps • Avg latency: 3ms • Hundreds of Stores • Largest store – 2.8TB+ Voldemort: Architecture
- 17. Kafka: High-Volume Low-Latency Messaging System LinkedIn Data Infrastructure Solutions 17
- 18. Kafka Architecture Producer Consumer Producer Consumer Zookeeper topic1-part1 topic2-part2 topic2-part1 topic1-part2 topic2-part2 topic2-part1 topic1-part1 topic1-part2 topic1-part1 topic1-part2 topic2-part2 topic2-part1 Broker 1 Broker 2 Broker 3 Broker 4 Key features • Scale-out architecture • High throughput • Automatic load balancing • Intra-cluster replication Per day stats • writes: 10+ billion messages • reads: 50+ billion messages
- 19. Filling in the Data Store Gap 19 Text Search
- 20. Espresso: Indexed Timeline-Consistent Distributed Data Store LinkedIn Data Infrastructure Solutions 20
- 21. Application View 21 Hierarchical data model Rich functionality on resources Conditional updates Partial updates Atomic counters Rich functionality within resource groups Transactions Secondary index Text search
- 22. Espresso: System Components 22 • Partitioning/replication • Timeline consistency • Change propagation
- 23. Generic Cluster Manager: Helix • Generic Distributed State Model • Config Management • Automatic Load Balancing • Fault tolerance • Cluster expansion and rebalancing • Espresso, Databus and Search • Open Source Apr 2012 • https://github.com/linkedin/helix 23
- 24. Infrastructure challenges in large-scale data mining Putting it together
- 25. Top complaints from data scientists 1 Getting the data in (Ingress ETL) 2 Getting the data out (Egress) 3 Workflow management 4 Model of computation 5 …
- 26. Top complaints from data scientists 1 Getting the data in (Ingress ETL) 2 Getting the data out (Egress) 3 Workflow management 4 Model of computation 5 …
- 27. LinkedIn circa 2010 LinkedIn Confidential ©2013 All Rights Reserved 27
- 28. O(n2) data integration complexity
- 29. Infrastructure fragility • Can‟t get all data • Hard to operate • Multi-hour delay • Labor intensive • Slow • Does it work?
- 30. Process fragility • Labor intensive • One man‟s cleaning… FE MT BE DT FE Dev BE Dev ETL Team ETL DW/ Hadoop
- 31. Data model { tracking_code=null, session_id=42, tracking_time=Tue Jul 31 07:27:25 PDT 2010, error_key=null, locale=en_us, browser_id=ddc61a81-5311-4859-be42-ca7dc7b941e3, member_id=1213, page_key=profile, tracking_info=Viewee=1214,lnl=f,nd=1,o=1214,^SP=pId- 'pro_stars',rslvd=t,vs=v,vid=1214,ps=EDU|EXP|SKIL|, error_id=null, page_type=FULL_PAGE, request_path=view ... }
- 32. Data model (cont‟d) { article_id=5560874437395353942, title=Five Good Reasons to Hire the Unemployed, language=en_US, article_source=bit.ly, url=aHR0cDovL3d3dy5vbmV0aGluZ25ldy5jb20vaW5kZXgucGhwL3dvcmsvMTAyLWZpdmUtZ29v ZC1yZWFzb25zLXRvLWhpcmUtdGhlLXVuZW1wbG95ZWQK, ... }
- 33. Problems 1 Data integration across systems 2 Fragile infrastructure 3 Lack of proper data models (ad-hoc)
- 34. LinkedIn 2013 LinkedIn Confidential ©2013 All Rights Reserved 34
- 37. Data model Hundreds of message types Thousands of fields What do they all mean? What happens when they change?
- 38. Data model 1 Education 2 Push data cleanliness upstream 3 O(1) ETL 4 Evidence-based correctness
- 39. Data model DDL for data definition and schema Central versioned registry of all schemas Schema review Programmatic compatibility model – Schema changes handled transparently
- 40. Workflow 1 Check in schema 2 Code review 3 Ship Seamless data load into downstream systems
- 41. Audit trail
- 42. Result: complete, verified copy of all data available
- 43. Top complaints from data scientists 1 Getting the data in (Ingress ETL) 2 Getting the data out (Egress) 3 Workflow management 4 Model of computation 5 …
- 44. Egress store DATA into „kafka://…‟ using Stream();
- 45. Top complaints from data scientists 1 Getting the data in (Ingress ETL) 2 Getting the data out (Egress) 3 Workflow management 4 Model of computation 5 …
- 47. Workflows 47 Job A Job B Job C Push to Production
- 48. Workflows 48 Job A Job B Job C Push to Production Job X
- 49. Workflows 49 Job A Job B Job C Push to Production Job X Push to QA
- 50. Real workflows are complicated 50
- 51. Workflow management: Azkaban 51 Dependency management Diverse job types (Pig, Hive, Java, . . . ) Scheduling Monitoring Configuration Retry/restart on failure Resource locking Log collection Historical information
- 54. Top complaints from data scientists 1 Getting the data in (Ingress ETL) 2 Getting the data out (Egress) 3 Workflow management 4 Model of computation 5 …
- 55. Model of computation • Alternating Direction Method of Multipliers (ADMM) • Distributed Conjugate Gradient Descent (DCGD) • Distributed L-BFGS • Bayesian Distributed Learning (BDL) Graphs Distributed learning Near-line processing
- 56. LinkedIn Data Infrastructure: A few take-aways LinkedIn Confidential ©2013 All Rights Reserved 56 1. Building infrastructure in a hyper-growth environment is challenging. 2. Few vs Many: Balance over-specialized (agile) vs generic efforts (leverage-able) platforms (*) 3. Balance open-source products with home- grown platforms (**) 4. Data Model and Integration e2e are key (*)
Editor's Notes
- #9: Transition needs to be goodProducts => data infrastructure requirements in previous slideAll products don’t make the same latency and freshness requirements from our data infrastructureThe way we bucketize this is….News and recommendations show up in both nearline and offline
- #40: Not part of kafka
- #52: - Others: Oozie
- #57: Data Integration is hard. Having sane and same metadata across systems. Have a schema which works across the 3 phases. Want a rich evolving schemas and make the conforming push as much of data cleaning to source and upstream as much as possible so near-line and off-line helpsSessionization logic is in WH which makes it hard for near-line systems to useExtensible system where changing schema in one phase does not break downstream systemsDon’t build over-specialized systems: e.g. a monitoring system for PYMK – build Azkaban