How Slack adopted Karpenter to increase Operational and Cost Efficiency (5 minute read)
Slack's internal Kubernetes platform, Bedrock, simplifies container deployment and management by using Amazon EKS and tools like Jenkins and Nebula overlay network. To enhance operational efficiency, Slack adopted Karpenter to dynamically provision nodes based on pod demands, leading to cost savings and faster scaling decisions, resulting in a 12% reduction in EKS compute costs.
It's always TCP_NODELAY. Every damn time (4 minute read)
Disabling Nagle's TCP_NODELAY socket option can lead to decreased network latency when sending outgoing packets on your server. This developer demonstrates that it's often worth disabling the default on option, particularly because it doesn't play well with TCP's delayed ACK feature.
Optimizing Clickhouse: 5 Tactics That Worked (17 minute read)
This article advocates for 5 tactics that helped Highlight.io improve their Clickhouse performance. The solutions include using larger insert batches, keeping data in wide parts (instead of compact), optimizing order by granularity, and checking merge levels.
Test Failures Should Be Actionable (2 minute read)
It's really important for a test failure to be actionable. Four guidelines that can help make tests actionable: write descriptive test names, keep tests focused on a specific scenario, narrow assertions in unit tests, and keep cause and effect clear.
Did GitHub Copilot really increase my productivity? (6 minute read)
This developer had free access to GitHub Copilot for a year, using it extensively for routine coding tasks. When the access ended, they reflected on its utility and concluded that despite its helpfulness with boilerplate code, they were more productive without it due to Copilot's unpredictability and slow response times. While initially beneficial for tedious code, the developer found the AI's erratic performance and delays disruptive, leading them to prefer coding without it.