Ingestion and Dimensions Compute and Enrich using Apache Apex
Download as PPTX, PDF5 likes693 views
Apache Apex is a platform for developing scalable, fault-tolerant distributed applications that process streaming and batch big data with low latency. The architecture supports real-time analytics through the integration of operators and data streams, optimizing resource usage and enabling efficient processing of large data volumes. Key features include dynamic partitioning, operational insight, and various operators for adapting to complex data processing needs.
![Dimensional Model
Dimensions Schema
{"keys":[{"name":"campaignId","type":"integer"},
{"name":"adId","type":"integer"},
{"name":"creativeId","type":"integer"},
{"name":"publisherId","type":"integer"},
{"name":"adOrderId","type":"integer"}],
"timeBuckets":["1h","1d"],
"values":
[{"name":"impressions","type":"integer","aggregators":["SUM"]},
{"name":"clicks","type":"integer","aggregators":["SUM"]},
{"name":"revenue","type":"integer"}],
"dimensions":
[{"combination":["campaignId","adId"]},
{"combination":["creativeId","campaignId"]},
{"combination":["campaignId"]},
{"combination":["publisherId","adOrderId","campaignId"],"additionalValues":["revenue:SUM"]}]
}](https://image.slidesharecdn.com/ingestanddimensions-160712192613/85/Ingestion-and-Dimensions-Compute-and-Enrich-using-Apache-Apex-29-320.jpg)
Ingestion and Dimensions Compute and Enrich using Apache Apex
- 1. Apache Apex Intro to Apex Ingestion and Dimensions Compute for a customer use-case Devendra Tagare [email protected] @devtagare 9h July 2016
- 2. What is Apex 2 • Platform and runtime engine that enables development of scalable and fault- tolerant distributed applications • Hadoop native • Process streaming or batch big data • High throughput and low latency • Library of commonly needed business logic • Write any custom business logic in your application
- 3. Applications on Apex 3 • Distributed processing • Application logic broken into components called operators that run in a distributed fashion across your cluster • Scalable • Operators can be scaled up or down at runtime according to the load and SLA • Fault tolerant • Automatically recover from node outages without having to reprocess from beginning • State is preserved • Long running applications • Operators • Use library to build applications quickly • Write your own in Java using the API • Operational insight – DataTorrent RTS • See how each operator is performing and even record data
- 5. Apex Operator Library - Malhar 5
- 6. Native Hadoop Integration 6 • YARN is the resource manager • HDFS used for storing any persistent state
- 7. Application Development Model 7 A Stream is a sequence of data tuples A typical Operator takes one or more input streams, performs computations & emits one or more output streams • Each Operator is YOUR custom business logic in java, or built-in operator from our open source library • Operator has many instances that run in parallel and each instance is single-threaded Directed Acyclic Graph (DAG) is made up of operators and streams Directed Acyclic Graph (DAG) Output StreamTuple Tuple er Operator er Operator er Operator er Operator er Operator er Operator
- 8. Advanced Windowing Support 8 Application window Sliding window and tumbling window Checkpoint window No artificial latency
- 10. Partitioning and unification 10 NxM PartitionsUnifier 0 1 2 3 Logical DAG 0 1 2 1 1 Unifier 1 20 Logical Diagram Physical Diagram with operator 1 with 3 partitions 0 Unifier 1a 1b 1c 2a 2b Unifier 3 Physical DAG with (1a, 1b, 1c) and (2a, 2b): No bottleneck Unifier Unifier0 1a 1b 1c 2a 2b Unifier 3 Physical DAG with (1a, 1b, 1c) and (2a, 2b): Bottleneck on intermediate Unifier
- 11. Advanced Partitioning 11 0 1a 1b 2 3 4Unifier Physical DAG 0 4 3a2a1a 1b 2b 3b Unifier Physical DAG with Parallel Partition Parallel Partition Container uopr uopr1 uopr2 uopr3 uopr4 uopr1 uopr2 uopr3 uopr4 dopr dopr doprunifier unifier unifier unifier Container Container NICNIC NICNIC NIC Container NIC Logical Plan Execution Plan, for N = 4; M = 1 Execution Plan, for N = 4; M = 1, K = 2 with cascading unifiers Cascading Unifiers 0 1 2 3 4 Logical DAG
- 12. Dynamic Partitioning 12 • Partitioning change while application is running ᵒ Change number of partitions at runtime based on stats ᵒ Determine initial number of partitions dynamically • Kafka operators scale according to number of kafka partitions ᵒ Supports re-distribution of state when number of partitions change ᵒ API for custom scaler or partitioner 2b 2c 3 2a 2d 1b 1a1a 2a 1b 2b 3 1a 2b 1b 2c 3b 2a 2d 3a Unifiers not shown
- 13. • Ingest from Kafka and S3 • Parse, Filter and Enrich • Dimensional compute for key performance indicators • Reporting of critical metrics around campaign monetization • Aggregate counters & reporting on top N metrics • Low latency querying using Kafka in pub-sub model Use Case ...
- 14. Screenshots - Demo UI
- 15. Proprietary and Confidential Scale • 6 geographically distributed data centers • Combination of co-located & AWS based DC’s. • > 5 PB under the data management • 22 TB / day data generated from auction & client logs • heterogeneous data log formats • North of 15 Bn impressions / day. • Average data inflow of 200K events/s 15
- 16. • Ad server log events consumed as Avro encoded, Snappy compressed files from S3. New files uploaded every 10-20 minutes. • Data may arrive in S3 out of order (time stamps). • Event size is about 2KB uncompressed, only subset of fields retrieved for aggregation. • Aggregates kept in memory (checkpointed) with expiration policy, query processing against in memory data. • Front-end integration through Kafka based query protocol for realtime dashboard components. Initial Requirements
- 17. Proprietary and Confidential Apache Apex 17 AdServer REST proxy REST proxy Real-time architecture- Powered By Apex Kafka Cluster S3Reader S3Reader Filter Operator Filter Operator Dimensions Aggregator Dimensions Aggregator Dimensions Store Query Query Result Kafka Cluster Auction Logs Middleware Auction Logs Filtered Events Filtered Events Aggregates Query from MW Query Query Results S3 S3 Client logsAuction Logs Architecture 1.0 - Batch Reads + Streaming Aggregations
- 18. • Unstable S3 client libraries – Unpredictable hangs and Corrupted data – On Hang, Master kills the container and restart reading of file from different container – Corrupt files caused containers to kill – application configurable retry mechanism and skip bad files – Limited read consumption throughput – 1 reader/file • Out of Order data – Some timestamp in future and past • Spike in load when new files are added followed by period of inactivity • Memory Requirement for Store – Cardinality Estimation for incoming data Challenges
- 19. Proprietary and Confidential Apache Apex 19 REST proxy Real-time architecture- Powered By Apex Client logs Kafka Input (Auction logs) ETL operator Filter Operator Filter Operator Dimensions Aggregator Dimensions Aggregator Dimensions Store/HDHT Query Query Result Kafka Cluster Auction Logs Kafka Cluster Middleware AdServer REST proxy Kafka Cluster Auction Logs Client logs Kafka Messages Decompress & Flatten Filtered Events Filtered Events Aggregates Query from MW Query Query Results S3 S3Reader Kafka Input (Auction logs) Auction Logs Architecture 2.0 - Batch + Streaming
- 20. Challenges • Complex Logical DAG • Kafka Operator Issues – Dynamic Partitioning – Memory Configuration – Offset snapshotting to ensure exactly once semantics • Resource Allocation – More memory requirement for Store (Large number of Unifiers) • Harder Debugging (More number of components) – GB(s) of container logs – Difficult to locate the sequence of failure • More of data transferred over wire within cluster • Limit Kafka read rate
- 21. Proprietary and Confidential 21 User Browser AdServer REST proxy REST proxy Real-time architecture- Powered By Apex Kafka Cluster Client logs Kafka Input (Auction logs) Kafka Input (Client logs) CDN (Caching of logs) ETL operator ETL operator Filter Operator Filter Operator Dimensions Aggregator Dimensions Aggregator Dimensions Store Query Query Result Kafka Cluster Auction Logs Client logs Middleware Auction Logs Client logs Kafka Messages Kafka Messages Decompress & Flatten Decompress & Flatten Filtered Events Filtered Events Aggregates Query from MW Query Query Results Architecture 3.0 - Streaming
- 23. Proprietary and Confidential Application Configuration • 64 Kafka Input operators reading from 6 geographically distributed DC’s • Under 40 seconds end to end latency - from ad-serving to visualization • 32 instances of in-memory distribute store • 64 aggregators • 1.2 TB memory footprint @ peak load • In-memory store was later replaced by HDHT for fault tolerance 23
- 24. Proprietary and Confidential Learning’s • DAG – sizing, locality & partitioning (Benchmark) • Memory sizing for the store or other memory heavy operators. • Cardinality estimation for incoming data is critical. • Upstream operators tend to require more memory than down-stream operators for high velocity reads. • Back pressure from down-stream failures due to skew in velocity of events & upstream failures .. Buffer Server sizing is critical. • For end to end exactly once its necessary to understand the external systems semantics & delivery guarantees. • Think fault tolerance & recovery before starting implementation. 24
- 25. Proprietary and Confidential Before And After 25 5 Hours + 20 Minute • No real-time processing system in place. • Publishers and buyers could only rely on a batch processing system for gathering relevant data • Outdated data, not relevant to current time • Current data being pushed to a waiting queue • Cumbersome batch-processing lifecycle • No visualization for reports • No glimpse into everyday happenings, translating to lost decisions or untimely decision making scenarios Before Scenario After Scenario • Phase 1 • With DataTorrent RTS (built on Apache Apex), Dev team put together the first real time analytics platform • This enabled Reporting of critical metrics around campaign monetization • Reuse of batch ingestion mechanism for the impression data, shared with other pipelines (S3) ~ 30 seconds No Real-time Batch + Real-time • Phase 2 • Reduce end-to-end latency through real-time ingestion of impression data from Kafka • Results available much sooner to the user • Balances load (no more batch ingestion spikes), reduces resource consumption • Handles ever growing traffic with more efficient resource utilization. Real-time Streaming
- 26. Proprietary and Confidential Operators used S3 reader (File Input Operator) • Recursively reading the contents of a S3 bucket based on a partitioning pattern • Inclusion & exclusion support • Fault tolerance (replay and idempotent) • Throughput of over 12K reads/second for event size of 1.2 KB each Kafka Input Operator • Ability to consume from multiple Kafka clusters • Offset management support • Fault tolerant reads • Support for idempotent & exactly once semantics • Controlled reads for managing back-pressure POJO Enrichment Operator • takes a POJO as input and does a look-up in a store for given key • supports caching • stores are pluggable • App builder ready 26
- 27. Proprietary and Confidential Operators used (cont …) Parser • Specify JSON schema • Emits a POJO based on the output schema • No user code required Dimension Store • Distributed in-memory store • Supports re-aggregation of events • Partitioning of aggregates per view • Low latency query support with a pub/sub model using Kafka HDHT • HDFS backed embedded key-value store • Fault tolerant, random read & write • Durability in-case of cold restarts 27
- 28. Dimensional Model - Key Concepts Metrics : pieces of information we want to collect statistics about. Dimensions : variables which can impact our measures. Combinations : set of dimensions for which one or metric would be aggregated.They are sub-sets of dimensions. Aggregations : the aggregate function eg.. SUM, TOPN, Standard deviation. Example : Dimensions - campaignId, advertiserId, time Metrics - Cost, revenue, clicks, impressions Aggregate functions -SUM,AM etc.. Combinations : 1. campaignId x time - cost,revenue 2. advertiser - revenue, impressions 3. campaignId x advertiser x time - revenue, clicks, impressions How to aggregate on the combinations ?
- 30. Proprietary and Confidential More Use-cases • Real-time Monitoring Alerts on deal tracking & monetization Campaign & deal health • Real-time Learning Using the lost bid insights for price recommendations. • Allocation Engine Feedback to ad serving for guaranteed delivery & line item pacing 30
- 31. Data Processing Pipeline Example App Builder 31
- 34. Real-Time Dashboards Real Time Visualization 34
- 35. Q&A 35
- 36. Resources 36 • http://apex.apache.org/ • Learn more: http://apex.apache.org/docs.html • Subscribe - http://apex.apache.org/community.html • Download - http://apex.apache.org/downloads.html • Follow @ApacheApex - https://twitter.com/apacheapex • Meetups – http://www.meetup.com/pro/apacheapex/ • More examples: https://github.com/DataTorrent/examples • Slideshare: http://www.slideshare.net/ApacheApex/presentations • https://www.youtube.com/results?search_query=apache+apex • Free Enterprise License for Startups - https://www.datatorrent.com/product/startup-accelerator/
Editor's Notes
- #27: Thomas – Mention these are extensions of malhar (Open Source)