2023/04/23 - Latest Engineering Blogs
This week’s newsletter collected 6 latest engineering blogs from LinkedIn, Uber, Etsy, Spotify, and Meta. You also can go to Tech Blog Search site to search engineering blogs from different IT companies.
[LinkedIn][2023/04/20] Viral spam content detection at LinkedIn
[Uber][2023/04/20] Measuring Performance for iOS Apps at Uber Scale
[Etsy][2023/04/18] How We Built a Multi-Task Canonical Ranker for Recommendations at Etsy
[LinkedIn][2023/04/18] Scaling Salt for Remote Execution to support LinkedIn Infra growth
[Spotify][2023/04/18] Fleet Management at Spotify (Part 1): Spotify’s Shift to a Fleet-First Mindset
[Meta][2023/04/17] A fine-grained network traffic analysis with Millisampler
[LinkedIn][2023/04/20] Viral spam content detection at LinkedIn
TL;DR
Problem. Viral spam content causes detrimental effects on user experience and engagement on LinkedIn. However, developing sophisticated models to fight viral spam content is challenging.
Solution. This blog proposed a hybrid strategy to fight viral spam content based on AI models, including proactive defense and reactive defense. Proactive defense filters out potential viral spam content as early as possible, while reactive defense filters out viral content based on continuous signals gathering.
[Uber][2023/04/20] Measuring Performance for iOS Apps at Uber Scale
TL;DR
Problem. Pre-warming in iOS 15 causes iOS app launch time (one of critical app performance metrics at Uber) measurement unreliable.
Solution. This blog redefined app startup latency, and proposed a new measurement pipeline to collect user level and aggregated level metrics at backend.
[Etsy][2023/04/18] How We Built a Multi-Task Canonical Ranker for Recommendations at Etsy
TL;DR
Problem. Etsy has hundreds of ranking models power stages of a user’s shopping journey. Each ranking model powers a specific objective, like finding trending items. The cost of maintaining and iterating on hundreds of ranking models is prohibitive.
Solution. This blog introduced a Multi-gate Mixture Model of Experts (MMOE) framework to optimize multiple objectives together, like users’ favorites and purchasing rate.
[LinkedIn][2023/04/18] Scaling Salt for Remote Execution to support LinkedIn Infra growth
TL;DR
Problem. LinkedIn’s engineers use Salt to execute tasks (e.g., OS upgrades) on hosts. With a high number of execution requests, Salt exhibits serious operational and performance challenges.
Solution. This blog proposed a new design to scale Salt for operational tasks to handle 10X more jobs with more reliability and scalability.
[Spotify][2023/04/18] Fleet Management at Spotify (Part 1): Spotify’s Shift to a Fleet-First Mindset
TL;DR
Problem. Major software updates (e.g., Java 17.0.4 to 17.0.5, Log4j vulnerability fix) are challenging and time consuming.
Solution. This blog proposed to build automation tools to safely make software updates/changes at once, and maintain the health of Spotify’s tech infrastructure continuously.
[Meta][2023/04/17] A fine-grained network traffic analysis with Millisampler
TL;DR
Problem. Meta needs a powerful tool to troubleshoot and analyze performance on network traffic faults in the data centers.
Solution. This blog presented Millisampler to observe, characterize, and debug network performance at high-granularity timescales efficiently.
If you have any questions or comments, please contact plussmart2018@gmail.com.