Gokul Raam — Today I Learnt

Deploying an full stack application

Mon, 01 Jun 2026 08:31:35 GMT

I mean, I have done it before, deploying full stack projects in college, as part of work regularly. But yeah, this time I'm doing it for my portfolio website. Doing it after long years. Will focus on load balancing, reverse proxy, security side soon!

Postgres EXPLAIN ANALYZE lies about buffers unless you ask

Thu, 28 May 2026 00:00:00 GMT

EXPLAIN ANALYZE shows timings but not I/O by default. To see what's actually being read from disk vs. cached, you need:

EXPLAIN (ANALYZE, BUFFERS, FORMAT TEXT)
SELECT * FROM users WHERE email = 'x@y.com';

The Buffers: line tells you shared hit=N read=N — that "read" is your real disk I/O. A query that's "fast" in EXPLAIN can still hammer the disk if you don't check this.

Pair with track_io_timing = on in postgresql.conf for actual I/O latency.

FastAPI Depends() doesn't cache across requests

Tue, 26 May 2026 00:00:00 GMT

Took me a minute. Depends() deduplicates within a single request — if two endpoints in a chain both depend on get_db(), it's resolved once. But the next request gets a fresh resolution.

That's a feature: per-request DB sessions are what you want. But if you expected app-wide singleton behavior (config, cached HTTP client), use a module-level instance or lru_cache:

from functools import lru_cache

@lru_cache
def get_settings() -> Settings:
    return Settings()

# Then in routes:
def handler(settings: Annotated[Settings, Depends(get_settings)]):
    ...

lru_cache makes get_settings a singleton; Depends happily reuses it.