now

Now

April 2026. Pune, via Udaipur.

Finishing the order placement service cutover at Repos Energy. The identity service migration is done — p99 is at 80ms and has been stable for two months. The order service is harder because its blast radius is larger and the traffic patterns are less predictable. Most of the remaining work is in chaos testing: I want a pprof baseline from production before and after the cutover so the comparison is real, not a number I have to reconstruct from memory.

Graduating from TIET in May. Four years of computer science, one IEEE paper on Zero-Trust network architecture, one merged contribution to a Google DeepMind codebase, and a clearer sense of which parts of this field I actually want to work in. The answer is: below the framework layer.

Reading Computer Systems: A Programmer's Perspective for the second time — the first time I read it for the content; this time I'm reading it for the pedagogy. How you explain a memory hierarchy constraint is almost as interesting as the constraint itself.

Writing: next post is about what pprof doesn't tell you — the gap between what the profiler surfaces and what's actually causing the problem. It's a harder post to write than the glibc one because the answer is less satisfying: sometimes the tool is right and you're looking at the wrong layer.

Updated: April 2026