Small Server Fit

Small Server Fit

Prevent avoidable deployment rework by checking application requirements against a measured server budget before committing to an image, runtime, or instance size.

Establish the two sides of the fit

Inspect the repository for:

• frontend build tool and whether static assets can be built locally or in CI;
• backend runtime and exact version constraints;
• lockfiles, native extensions, browser binaries, machine-learning model artifacts, and serializer versions;
• steady processes, intended worker count, background jobs, databases, uploads, logs, and temporary build space;
• representative validation flow and concurrency expectation.

Obtain a current or candidate server profile containing OS image, architecture, vCPU, RAM, swap, disk and free space, available runtime versions, and platform services. Use references/profile-schema.md for the evidence contract.

If the host does not exist yet, mark runtime measurements NOT MEASURED and separate image compatibility from capacity claims. Do not treat a purchase date or marketing label as proof of installed runtime versions.

Make the fit decisions

Evaluate these independently:

1. Operating-system and architecture fit: application binaries and native packages support the image and architecture.
2. Runtime fit: an application-specific Python, Node.js, or Java runtime can be installed without replacing a system-owned runtime.
3. Build fit: peak dependency installation and build memory and disk fit, or the build can move to a local machine or CI.
4. Runtime memory fit: measured or bounded steady processes leave headroom for the OS, Nginx, SSH, logs, and short spikes.
5. Persistence fit: state is outside immutable release directories and backups fit the disk or an external store.
6. Operations fit: the service can restart, recover after reboot, expose only intended ports, and produce bounded logs.

Run scripts/evaluate_fit.py --profile <profile.json> when numeric evidence is available. The script calculates margins; it does not invent missing workload measurements.

Prefer changes in this order

1. Build the SPA or Java artifact outside the small server.
2. Isolate the application runtime instead of upgrading system binaries.
3. Reduce the process or worker count when the workload and state model permit it.
4. Move databases, object storage, or heavy build jobs to a managed or external service when justified.
5. Add swap only as a bounded resilience measure, not as a substitute for required RAM.
6. Increase the instance size when measured steady-state plus safety margin still does not fit.

Never recommend downgrading security, removing backups, disabling verification, or co-locating incompatible state merely to claim fit.

Output

Report each dimension as FIT, FIT WITH CHANGES, NOT FIT, or NOT MEASURED. Include:

• the recommended OS image and runtime isolation method;
• local or CI versus server build split;
• process and worker budget;
• RAM and disk arithmetic with source measurements;
• model and native dependency compatibility risks;
• the smallest safe next measurement or experiment;
• a purchase recommendation only when evidence supports it.

Distinguish a confirmed fact, calculation, assumption, and recommendation. A useful result can be NOT MEASURED; false precision cannot.