Cutting My Self-Hosted Analytics Stack Memory by 45%

260MB for a personal analytics dashboard. Just tracking 4,000 events across 21 websites over 5 days. Barely any data, but my Bun server was eating 150MB and TimescaleDB was claiming 120MB like it was serving a production SaaS.

On Railway's usage-based $5 plan, every MB counts. Time to optimize.

The Stack

• Server: Bun + Hono + Drizzle ORM + Better-Auth
• Database: TimescaleDB (PostgreSQL for time-series data)
• Cache: Redis
• Frontend + Proxy: Nginx serving Vite/React SPA with reverse proxy to backend

Server Optimizations (150MB → 50-60MB)

1. Removed unused OpenAPI/Zod schema introspection - heavy libs not needed for single-user project
2. Used Bun's --smol flag - smaller memory footprint GC with more aggressive garbage collection. Minimal perf impact for low-traffic apps. This one's a hidden gem.
3. Removed --sourcemap and --bytecode - unnecessary overhead in production binaries

TimescaleDB Optimizations (100-120MB → 60-80MB)

The TimescaleDB defaults were completely overkill for a personal project:

Setting	Default	Optimized	What it does
shared_buffers	1907MB	128MB	Memory for caching table/index data. Default was set to 25% of RAM - insane for a personal project
maintenance_work_mem	953MB	64MB	Memory for VACUUM, CREATE INDEX, etc. Almost 1GB for maintenance ops!
work_mem	2.4MB	4MB	Memory per query operation (sorts, joins). This was already reasonable
max_background_workers	16	8	Workers for chunk compression/maintenance. Halved but still plenty
telemetry_level	basic	off	Disables TimescaleDB telemetry. Saves CPU cycles and network calls

-- Memory settings (tuned down for single-user workload)
ALTER SYSTEM SET shared_buffers = '128MB';
ALTER SYSTEM SET maintenance_work_mem = '64MB';
ALTER SYSTEM SET work_mem = '4MB';

-- TimescaleDB settings
ALTER SYSTEM SET timescaledb.max_background_workers = '8';
ALTER SYSTEM SET timescaledb.telemetry_level = 'off';

SELECT pg_reload_conf();

Results

Component	Before	After	Reduction
Bun Server	150MB	55MB	-63%
TimescaleDB	110MB	70MB	-36%
Redis	8MB	8MB	Already lean
Nginx + Frontend	35MB	35MB	Already lean
Total	~300MB	~170MB	-45%

From a memory-hungry pet project to something a Raspberry Pi could handle without breaking a sweat.

Key Takeaway

Don't assume defaults are right for your use case. A personal project doesn't need the same configuration as a high-traffic SaaS. Right-size your infrastructure - that 130MB headroom means I can spin up another test project or two without bumping tiers.