Every note starts as a question I hit in real work. I research it until the answer holds up, then write the version I wish I had found. Sources are listed at the end of each one.
9 notes shown
Wall clocks drift, step and freeze, so created_at cannot order writes. How last-write-wins loses data, and what logical clocks and fencing tokens fix.
The path from await query() to a row has five measurable stages. Which one owns your latency is decided by deployment shape, not by the query.
tRPC's maintainer recommends GraphQL or OpenAPI once ownership crosses a team boundary. What that boundary costs in each of the three.
highWaterMark gates one boolean and buffers the chunk anyway. What Node's write buffer really is, why drain fires once, and where pipe() leaks.
From v27 every Node major becomes LTS and odd releases disappear. What that changes in a CI matrix, why the 30-month window survived, and what it misses.
Scale-to-zero rarely reaches zero. How to compute the serverless-versus-provisioned Postgres crossover from your workload's duty cycle.
Postgres biases its own defaults toward your index. When it still picks a sequential scan, the cost model is telling you what it cannot measure.
Why 'which runtime is faster' is the wrong question — what benchmark methodology hides, and how to actually pick between Node, Bun, and Deno.
I build web platforms the same way I write these notes: find out how it actually works, then make the call you can defend six months later.
Start a project