Accumulo Summit 2016: Timely - Scalable Secure Time Series Database
0 likes308 views
Timely is a scalable and secure time series database built on Apache Accumulo, designed for high performance and reliability. It integrates with Grafana for data visualization and offers support for various communication protocols and security configurations. The system allows for flexible data ingestion and subscription, with documented functionalities for managing time series data efficiently.
![Timely Input Format (Text)
lSimple text based on OpenTSDB put format:
put <metric> <timestamp> <value> <tag>[,<tag>...]
lExample
put sys.cpu.idle 1469735914000 25.0 host=s01n04 rack=s01 instance=0
lSupported in all protocols
lviz tag used to label data
– viz=private](https://image.slidesharecdn.com/timely-161012164045/85/Accumulo-Summit-2016-Timely-Scalable-Secure-Time-Series-Database-6-320.jpg)
Accumulo Summit 2016: Timely - Scalable Secure Time Series Database
- 1. Scalable Secure Time Series Database https://NationalSecurityAgency.github.io/timely
- 2. Overview lBuilt on Apache Accumulo – Proven Security, Scale & Reliability lUses Netty for communication protocols – Widely adopted, easy to integrate lProvides secure access to labeled data – Easily customized to meet unique architectures
- 3. History lIntegrated OpenTSDB with Apache Accumulo – Using Eric Newtons shim code – Seemed to have issues with scale – FAIL - Could not get past StackOverflowError •(OpenTSDB issue #334) lDecided to write it from scratch – Keep Grafana – Use Grafana OpenTSDB datasource plugin lHad something working in 2 weeks
- 4. Simple Architecture lInsert data points lSubscribe to data points lQuery for aggregated data points Timely Ingest Subscribe Time Series
- 5. Application Interfaces lSupports multiple protocols – udp, tcp, https, websocket lOperations for storing data – All protocols, security tag optional lOperations for working with time series data – https and websocket lOperations for subscribing to data – websocket only
- 6. Timely Input Format (Text) lSimple text based on OpenTSDB put format: put <metric> <timestamp> <value> <tag>[,<tag>...] lExample put sys.cpu.idle 1469735914000 25.0 host=s01n04 rack=s01 instance=0 lSupported in all protocols lviz tag used to label data – viz=private
- 7. Timely Input Format (Binary) lBinary format uses Google FlatBuffers encoding lIDL file located in the source code lGenerate client code in multiple languages lCurrently supported in UDP and TCP protocols
- 8. Sending Data to Timely lSend data directly from your application lCan use existing collection agents: – OpenTSDB Tcollector – CollectD lCan leverage StatsD servers also – HADOOP-12360 (StatsD Metrics2 sink)
- 9. Storage Format lMeta Table – Stores unique metric and tag information lMetrics Table – Stores individual metric data – Each data point stored N ways, N = # tags lSeveral bytes to store each key – Run Length Encoding – Compression
- 10. Visualizing Time Series Data lTimely built to work with Grafana lTimely App for Grafana – Drop it into the Grafana plugins directory – Provides Timely data source – Integrates security features into Grafana – Example dashboards provided
- 11. Timely App – Data Sources lDefine Timely Data Sources lTest Connectivity
- 12. Timely App – Menu Items •Login to defined data source lView Metric Names / Tags
- 13. Timely App – Login lTop – Login using client certificates lBottom – Login using username / password
- 14. Sample Dashboards lTimely App included dashboards: – Timely Status – System Overview – Hadoop Overview – Accumulo Overview
- 15. System Overview
- 19. HDFS DataNode Metrics (cont.)
- 22. Subscribing to Data lSubscription API over WebSocket protocol – WebSocket is a bi-directional protocol – Timely uses secure WebSockets (wss) lCreate connection and subscribe to: – Data for specific metric names – Data for a specific time window – Optionally, data that matches tag names and values lCan register multiple subscriptions lRemove subscriptions when appropriate
- 23. Security - Implementation lTimely stores the labels provided in the viz tag – Timely only calls flatten() on the CV for consistent ordering lSpring Security enables users to plug in their authentication mechanism and role provider lWorkflow: – User logs into Timely via /login HTTPS endpoint – User authenticated via Spring Security – HTTP secure session cookie returned for future API calls
- 24. Security Configuration lAnonymous access configurable lSSL provider: JDK or OpenSSL lSSL file locations and passwords lSSL ciphers lSession cookie expiration lCORS properties
- 25. Transport Security lHTTP Strict Transport Security (HSTS) – Accessing via http will redirect to HTTPS – Rule stored in browser for configured time lHTTPS lWSS
- 26. Modes of Operation lAnonymous access enabled – Unauthenticated users only see unlabled data – Authenticated users see what they are allowed lAnonymous access disabled – Unauthenticated users receive an error message – Authenticated users see what they are allowed
- 27. Roadmap lSummarization of historical data lNew Time Series API – Move away from OpenTSDB API – Add additional features lTimely Client – Make subscribing to data easier – Enable analytics to be easily written lEnrichment – Allow for user supplied information about time series lSupport Grafana annotations
- 28. Deploying Timely lJava 8 required for Accumulo and Timely lTested with Accumulo 1.7.x and Hadoop 2.6 lStandaloneMode – Uses Mini Accumulo Cluster – Useful for development and testing – Data lost across restarts lNon-Standalone Mode – 1+ Timely Servers
- 29. Deployment #1 lSetup: – 1 Timely Server – Accumulo 1.7.1, 26 Tservers on single disk hosts lTimely server receiving 2.75M metrics/min l Inserting 20.3M keys/min (338K / sec) – @10:1 ratio inserted to received l2.2T keys in the metrics table – 8.75TB unreplicated – @ 4.3 bytes per key, ~ 40 bytes per metric
- 30. Deployments #2 lSetup: – 2 Timely servers – Accumulo 1.7.1, 31 TabletServers on single disk hosts lTimely servers receiving 10M metrics/minute lInserting 71M keys/minute (1.18M / sec) – @ 7:1 ratio inserted to received l1.91T keys in the metrics table – 7.47TB unreplicated – @4.3 bytes per key, ~ 30 bytes per metric
- 31. Questions?