polite-http

A courteous, dependency-free HTTP client for Python. It plays nice with rate-limited APIs out of the box:

• Per-host rate limiting — cross-process, via a shared file lock.
• Automatic retries on transient errors (HTTP 429, 5xx) and network errors.
• Exponential backoff with optional jitter.
• Retry-After support (server-directed backoff, both seconds and HTTP-date forms).
• X-Throttling-Control proactive backpressure (as used by PubChem / NCBI).
• Streaming helpers for large line-oriented and binary responses.
• Zero third-party dependencies — built entirely on the standard library (urllib.request).

Installation

pip install polite-http

Requires Python 3.9+. Cross-process rate limiting works on Linux, macOS, and Windows — it uses fcntl on POSIX and msvcrt on Windows for the shared file lock (both standard library). On the rare platform that provides neither, it falls back to a best-effort in-process timer.

Quick start

from polite_http import HttpClient

# Scope a client to a base URL and a steady-state rate of 3 requests/second.
client = HttpClient("https://eutils.ncbi.nlm.nih.gov/entrez/eutils/", qps=3)

# GET + parse JSON (relative paths are resolved against the base URL).
data = client.fetch_json("esummary.fcgi?db=pubmed&id=123456")

# POST a JSON body.
result = client.fetch_json(
    "esearch.fcgi",
    method="POST",
    json_body={"db": "pubmed", "term": "cancer"},
)

# Download raw bytes with a custom timeout.
pdf = client.fetch_bytes(
    "efetch.fcgi?db=pubmed&id=123456&rettype=abstract",
    timeout=60,
)

Streaming

# Stream a large response line-by-line without buffering it all in memory.
for line in client.stream_lines("large-export.tsv"):
    process(line)

# Stream binary content in chunks (e.g. to a file).
with open("paper.pdf", "wb") as f:
    for chunk in client.stream_bytes("paper.pdf"):
        f.write(chunk)

Configuration

HttpClient accepts the following keyword arguments:

Environment variables

• POLITE_HTTP_USER_AGENT — default User-Agent when one isn't passed explicitly. Many APIs (e.g. NCBI) reject requests without a descriptive User-Agent, so setting this is recommended.
• POLITE_HTTP_LOCK_DIR — directory for the cross-process rate-limit lock files (defaults to the system temp directory).

Error handling

Failed requests raise HttpError, which carries the status_code, raw body bytes, and url:

from polite_http import HttpClient, HttpError

client = HttpClient("https://api.example.com/", qps=5)
try:
    data = client.fetch_json("widgets/42")
except HttpError as exc:
    print(exc.status_code, exc.url)
    if exc.body:
        print(exc.json())  # parse the error body as JSON, if applicable

Acknowledgements

The HTTP client at the heart of this package is derived from the science-skills project by Google DeepMind, used under the Apache License 2.0. See NOTICE for details of the changes.

License

Apache License 2.0.