Files, Config, and Templates¶

By Sameer Alam Updated 14 September 2026

What You'll Learn¶

How to work with paths and files safely using pathlib
How to read and write JSON, YAML, and TOML — and the YAML pitfalls to avoid
How to layer defaults, environment files, and environment variables into one config
How to render config files with Jinja2 and write them atomically

`pathlib`¶

from pathlib import Path

base = Path("/etc/opsctl")
config_file = base / "config.yaml"          # join with /

config_file.exists()
config_file.is_file()
config_file.parent                          # /etc/opsctl
config_file.suffix                          # .yaml
config_file.stem                            # config

text = config_file.read_text(encoding="utf-8")
Path("out/report.txt").parent.mkdir(parents=True, exist_ok=True)
Path("out/report.txt").write_text("done\n", encoding="utf-8")

for log in Path("/var/log/app").glob("*.log"):            # one directory
    print(log.name, log.stat().st_size)
for manifest in Path("k8s").rglob("*.yaml"):              # recursive
    print(manifest)

home_config = Path.home() / ".config" / "opsctl" / "config.yaml"

Always pass encoding="utf-8" explicitly; the default depends on the platform.

JSON¶

import json

data = json.loads(Path("inventory.json").read_text(encoding="utf-8"))
Path("inventory.json").write_text(json.dumps(data, indent=2, sort_keys=True) + "\n", encoding="utf-8")

# Newline-delimited JSON logs: one object per line
with Path("app.json.log").open(encoding="utf-8") as fh:
    errors = [rec for line in fh if (rec := json.loads(line)).get("level") == "ERROR"]

Datetimes aren't JSON serializable. Convert them explicitly: json.dumps(obj, default=str), or format with .isoformat().

YAML¶

import yaml

with Path("values.yaml").open(encoding="utf-8") as fh:
    values = yaml.safe_load(fh)

with Path("out.yaml").open("w", encoding="utf-8") as fh:
    yaml.safe_dump(values, fh, sort_keys=False, default_flow_style=False)

# Multi-document files, such as Kubernetes manifests
with Path("manifests.yaml").open(encoding="utf-8") as fh:
    for doc in yaml.safe_load_all(fh):
        if doc:
            print(doc["kind"], doc["metadata"]["name"])

Always safe_load

yaml.load without a safe loader can construct arbitrary Python objects from tags in the file — code execution from a config file. Use yaml.safe_load and yaml.safe_dump.

YAML surprises¶

country: NO          # PyYAML (YAML 1.1) reads this as False
version: 1.10        # a float: 1.1
port: 0800           # may parse as an octal integer, or a string
enabled: on          # True in YAML 1.1

Quote values that must stay strings: country: "NO", version: "1.10". PyYAML also drops comments on rewrite; when you must edit a human-maintained YAML file and keep its comments and ordering, use ruamel.yaml.

TOML¶

Python 3.11+ reads TOML in the standard library:

import tomllib

with Path("pyproject.toml").open("rb") as fh:     # tomllib needs binary mode
    project = tomllib.load(fh)
print(project["project"]["version"])

To write TOML, use the tomli-w package.

Layered Configuration¶

Real tools combine several sources. A clear, predictable order:

flowchart LR
  A["Built-in defaults"] --> B["config file"]
  B --> C["environment-specific file"]
  C --> D["environment variables"]
  D --> E["command-line flags"]
  E --> F["Validated config object"]

src/opsctl/config.py

import os
from dataclasses import dataclass, field
from pathlib import Path

import yaml

MAX_TIMEOUT_SECONDS = 120

DEFAULTS = {
    "region": "us-east-1",
    "timeout_seconds": 10,
    "health_endpoints": [],
    "notify": {"slack_channel": None},
}


def deep_merge(base: dict, override: dict) -> dict:
    merged = dict(base)
    for key, value in override.items():
        if isinstance(value, dict) and isinstance(merged.get(key), dict):
            merged[key] = deep_merge(merged[key], value)
        else:
            merged[key] = value
    return merged


@dataclass(frozen=True)
class Config:
    environment: str
    region: str
    timeout_seconds: int
    health_endpoints: list[str] = field(default_factory=list)
    slack_channel: str | None = None

    def __post_init__(self) -> None:
        if self.environment not in {"dev", "staging", "prod"}:
            raise ValueError(f"unknown environment: {self.environment!r}")
        if not 1 <= self.timeout_seconds <= MAX_TIMEOUT_SECONDS:
            raise ValueError(f"timeout_seconds must be between 1 and {MAX_TIMEOUT_SECONDS}")


def load_config(environment: str, config_dir: Path = Path("config")) -> Config:
    data = DEFAULTS
    for name in ("base.yaml", f"{environment}.yaml"):
        path = config_dir / name
        if path.exists():
            data = deep_merge(data, yaml.safe_load(path.read_text(encoding="utf-8")) or {})

    # Environment variables win over files
    if region := os.environ.get("OPSCTL_REGION"):
        data["region"] = region
    if timeout := os.environ.get("OPSCTL_TIMEOUT_SECONDS"):
        data["timeout_seconds"] = int(timeout)

    return Config(
        environment=environment,
        region=data["region"],
        timeout_seconds=data["timeout_seconds"],
        health_endpoints=list(data["health_endpoints"]),
        slack_channel=data["notify"]["slack_channel"],
    )

Validating in one place means a typo like timeout_seconds: "ten" fails at startup with a clear message, not halfway through a run. For larger schemas, pydantic provides validation, type coercion, and good error messages.

Keep secrets out of config files. Reference them by name and fetch them at runtime from a secrets manager or environment variables injected by the platform.

Rendering Config Files With Jinja2¶

templates/nginx-upstream.conf.j2

# Managed by opsctl — do not edit by hand
upstream {{ service }} {
    least_conn;
{% for backend in backends %}
    server {{ backend.host }}:{{ backend.port }}{% if backend.weight %} weight={{ backend.weight }}{% endif %};
{% endfor %}
    keepalive {{ keepalive | default(32) }};
}

from jinja2 import Environment, FileSystemLoader, StrictUndefined

env = Environment(
    loader=FileSystemLoader("templates"),
    undefined=StrictUndefined,      # a missing variable is an error, not an empty string
    trim_blocks=True,
    lstrip_blocks=True,
    keep_trailing_newline=True,
    autoescape=False,               # config files, not HTML
)

rendered = env.get_template("nginx-upstream.conf.j2").render(
    service="orders_api",
    backends=[{"host": "10.0.2.15", "port": 8080}, {"host": "10.0.2.16", "port": 8080, "weight": 2}],
)

Without StrictUndefined, a misspelled variable renders as an empty string and produces a config file that looks valid but isn't. The same templating language is used by Ansible templates.

Atomic Writes¶

If a process crashes halfway through writing a config file, readers can see a truncated file. Write to a temporary file in the same directory, then rename it into place — renames are atomic on the same filesystem:

import os
import tempfile
from pathlib import Path

def atomic_write(path: Path, content: str, mode: int = 0o644) -> None:
    path.parent.mkdir(parents=True, exist_ok=True)
    fd, tmp = tempfile.mkstemp(dir=path.parent, prefix=f".{path.name}.")
    try:
        with os.fdopen(fd, "w", encoding="utf-8") as fh:
            fh.write(content)
            fh.flush()
            os.fsync(fh.fileno())
        os.chmod(tmp, mode)
        os.replace(tmp, path)
    except BaseException:
        Path(tmp).unlink(missing_ok=True)
        raise

Only rewrite when content changes, so file watchers and reloads don't trigger needlessly:

target = Path("/etc/nginx/conf.d/orders-upstream.conf")
if not target.exists() or target.read_text(encoding="utf-8") != rendered:
    atomic_write(target, rendered)
    subprocess.run(["nginx", "-t"], check=True, timeout=30)
    subprocess.run(["systemctl", "reload", "nginx"], check=True, timeout=30)

Common Mistakes¶

yaml.load instead of yaml.safe_load.
Unquoted YAML values like NO, on, or 1.10 changing type silently.
Rendering templates without StrictUndefined, shipping configs with empty values.
Writing config files in place, so a crash or full disk leaves a truncated file.
Scattering os.environ lookups throughout the code instead of loading and validating config once.
Putting secrets in YAML config files committed to Git.

Interview Questions¶

Why is yaml.safe_load important?
How would you layer defaults, per-environment files, and environment variables into one configuration?
Why use StrictUndefined in Jinja2?
How do you update a config file so that readers never see a partially written file?

Next¶

Continue to Testing, Linting, and Packaging.