Yuqi Zhang

CPC Resiliency 𝑺ituation: Ads real time pipeline crashed too frequently even in “Black Friday”. As a result, seller campaigns are stopped too late, which causes the Ads revenue loss. 𝑻 ask: Find out all parts that have single point failure risks in the whole pipeline and fix them. 𝑨ction: Work together with Sojourner, Rheos, and Attribution team, integrate Kafka HA topic, use BES as the backup stream, build E2E monitor and design a two stream auto switch architecture. 𝑹esult: Improve the high availabilities from 90% to 99.999%. Save 5 million revenue. Achievements are shown in eBay CCOE newsletter and acknowledged by VP.

Rule Engine 𝑺ituation: We need a rule engine to support click fraud detection for products like PLP, SFA and PD. 𝑻 ask: The rule should be written once and executed both in Spark and Flink. 𝑨ction: Develop a service for storing business rules. Design and implement the rule expression language for both Spark and Flink. Rewriting Flink stream group operator. 𝑹esult: Accelerate the velocity of developing new rules. Improve the Spark job performance by reducing the run time from 2.5h to 30min.

Omni Search Microsoft, Cloud & AI team, SCM group, Intern 𝑺ituation: Dynamics AX and Outlook plugin both need a Search Service to provide search ability. 𝑻ask: Build a Search service and web service to extract key words intelligently from Outlook Email and automatically show potential results that the user wants. 𝑨ction: Build a web form for administrator to decide which tables and columns are going to be indexed. Developing and scheduling batch jobs to update the revert index. 𝑹esult: Support search requests from Dynamics AX and Outlook plugin. Improve the performance of updating index from 30 seconds within 5 second.