Month 9 (1?)

This is my brain dump for this month on things I learned, tech-related or otherwise. I’m shifting from a weekly to a realistic monthly blog :)

Tech

• I dug into , the go logging library, aptly named so. It turns out to be 4-10x faster than most alternatives. It provides two main apis
  • one Logger with single method for each level (Info(), Warn(), Error())
  • second, higher level api called SugaredLogger which has a less verbose api at a small cost to the performance. Worth it for everyday dev work, but probably not in a hot path.
    • convert to second using .Sugar()
• Frequent reauth is not the solution for a secure web experience. Having short session timeouts (unless for specific highly-secure use cases like bank transactions) are far too long to deter session hacking and too short to cause login fatigue. This leads to reused passwords and insecure methods of login, counterproductive to the cause. The solution is continuous verification rather than reauth.
• Bloom Filters were used by Akamai Tech to prevent one hit wonders from clogging up their caching systems. Using a bloom filter to add the first request, and only caching the object on the second request if the bloom filter is positive is a optimization that is easy but effective
• Coinbase spent $65 million with datadog for their observability tooling. Other options for migration included prometheus and grafana, but the development effort trumped the ease of dev experience with datadog. Apparently, the money and expertise saved by auto monitoring is worth it, in the long haul.
• The windows option to toggle off seconds rendering on the clock might have actual battery implications according to tests. This could be due to the device having to render every second rather than minute deterring the CPU from ever entering low power states
• I have been working with Elastic Kubernetes Services at Amazon and recently, EKS announced support for 100000 node clusters. This, along with generation accelerated ec2 instances would mean running 800k NVIDIA GPUs in a single cluster (that blows my mind). Unlocks crazy things with relation to LLMS and genai workloads
• Every android phone now comes with the accelerometers, small devices that enable the phone to sense changes in motion, mainly used for counting steps, but there is a secondary use for this sensor. The emergency alert system aggregates this data and sends out earthquake alerts if MANY phones in the area see concurrent vibrations, this is triggered only an alert when there’s a widespread phone activity that matches the pattern expected of an earthquake

  otherwise

• Nothing to report, have a good day reader