Microservices Practitioner Summit Jan '15 - Scaling Uber from 1 to 100s of Services - Matt Ranney, Uber
9 likes•6,196 views
As of January 2016, Uber operates in 361 cities across 67 countries with 5,400 employees and 1,700 engineers, generating $3.5 billion in driver payments from January to October 2015. The document details Uber's engineering evolution from 2009 to 2016, highlighting significant technological shifts and challenges in microservices, scalability, and latency issues. It emphasizes the ongoing technical debt and attempts to improve system performance through various programming languages and architectural strategies.
Microservices Practitioner Summit Jan '15 - Scaling Uber from 1 to 100s of Services - Matt Ranney, Uber
- 2. As of January 2016: Uber Cities Worldwide: 361 Countries: 67 Employees: 5,400 Engineers: 1,700 US Driver Payments Jan-Oct 2015: $3.5B
- 6. 2009-2010 Outsourced PHP + MySQL Jan 2011 "dispatch" - Node.JS/MongoDB Jan 2011 “API” - Python/SQLAlchemy/MySQL Feb 2012 Dispatch swaps MongoDB for Redis May 2012 Dispatch adds ON fallback Jan 2013 First non-API Python services Feb 2013 API switched to Postgres Mar 2014 New Python services use MySQL Mar 2014 Schemaless begins, must finish before pg collapse Sep 2014 First Schemaless - trips out of Postgres Aug 2015 Dispatch X.0 / Ringpop / Riak Jan 2016 Go, Java, Cloud, More Abstractions
- 11. Credit: NASA, ESA, and R. Thompson (Univ. Arizona)
- 12. Credit: NASA, ESA, and Z. Levay (STScI/AURA)
- 15. SCALING NODE
- 37. Getting out of the HTTP+JSON business HTTP is slow, complex, and inconsistent JSON is hard to validate, awkward in non-node Thrift is OK, but generated code is bad
- 42. Lots of services, lots of instances Mostly Node.JS and Python Call graph unknowable Self-inflicted DoS Cascading failures
- 48. load balancerservice A service B service B
- 49. load balancer service A service B service B
- 51. horizontal scalability zipkin tracing circuit breaking rate limiting failure testable almost no configuration as available as possible
- 52. overall latency ≥ latency of slowest component 1ms avg, 1000ms p99 use 1: 1% at least 1000ms use 100: 63% at least 1000ms 1.0 - 0.99^100 = 0.634 = 63.4% LATENCY
- 53. requeststhatareslow 0% 25% 50% 75% 100% Processes Used 1 2 4 8 16 32 64 128 256 512 1024 p95 p99 p99.9
- 57. CULTURALCHANGES
- 58. FAILURE TESTING
- 59. RETRIES
- 60. partner app dispatch DC1 Location Updates State Digest dispatch DC2 Location UpdatesState Request
- 61. THANKS