web_scraper/README.md

# Web Scraper — Book Catalogue

A production-style Python scraper for **books.toscrape.com**, a public
e-commerce-style catalogue built specifically for scraping practice. It
downloads the full 1,000-book catalogue across 50 paginated pages and writes
a clean, typed CSV — and can optionally enrich each row by visiting the
individual product page.

## What it does

| Field | Source | Notes |
|---|---|---|
| `title` | `h3 a[title]` | The visible text is truncated; the full title lives in the `title` attribute |
| `upc` | listing card | Optional; empty when the site omits it |
| `price_gbp` | `p.price_color` | `£51.77` → `51.77` (float) |
| `rating_1to5` | star-rating classes | `['star-rating','Three']` → `3` |
| `availability` / `num_available` | `p.instock.availability` | text + parsed stock count |
| `category` | product-page breadcrumb | not on the listing — fetched per book with `--detail` |
| `product_url` | `h3 a[href]` | resolved to an absolute URL |
| `scraped_at` | run time | UTC ISO-8601 |

Real run on 2026-09-26: **1,000 books, 50 categories, ~1m43s end-to-end**
(including 1,000 detail-page fetches at 6 workers). The committed
`books.csv` is the actual output of that run — not a sample.

## Install

```bash
pip install -r requirements.txt
```

## Run

```bash
# Catalogue only (fast, ~40s for all 50 pages)
python scraper.py books.csv

# Catalogue + per-product category enrichment (slower, ~2min)
python scraper.py books.csv --detail --workers 6

# Just look at the first 2 pages, with logging
python scraper.py sample.csv --max-pages 2 --verbose
```

## Sample output (first rows of the real `books.csv`)

```
title,upc,price_gbp,rating_1to5,availability,num_available,category,product_url,scraped_at
A Light in the Attic,,51.77,3,In stock,0,Poetry,https://books.toscrape.com/catalogue/a-light-in-the-attic_1000/index.html,2026-09-26T09:42:40Z
Tipping the Velvet,,53.74,1,In stock,0,Historical Fiction,https://books.toscrape.com/catalogue/tipping-the-velvet_999/index.html,2026-09-26T09:42:40Z
Soumission,,50.1,1,In stock,0,Fiction,https://books.toscrape.com/catalogue/soumission_998/index.html,2026-09-26T09:42:40Z
```

## Implementation details worth knowing

* **Retries with exponential backoff** — `fetch()` retries on network errors
  and 5xx responses (4 attempts: 1.5s / 2.25s / 3.4s waits).
* **Encoding pitfall, fixed** — the server sends `Content-Type: text/html`
  with no charset, so `requests` falls back to ISO-8859-1 and the `£` sign
  becomes mojibake. `fetch()` pins `resp.encoding = "utf-8"` when the charset
  is missing or latin-1. This is the single most common silent bug in
  scrapers; it's handled explicitly.
* **Pagination is link-driven** — it follows the `li.next a` href until it
  disappears, so adding/removing pages doesn't break the run.
* **Polite by default** — a `requests.Session` reuses the TCP connection, a
  custom User-Agent is set, and there's a configurable delay between pages.

## How I'd scale this

* **Rate limiting / politeness:** move the fixed `POLITE_DELAY` to a token
  bucket so bursts are capped but throughput isn't needlessly throttled;
  honor `Retry-After` headers when the site returns 429.
* **Retries:** swap the hand-rolled backoff for `urllib3`/`tenacity` with
  jitter (full-jitter avoids thundering-herd when many workers recover at
  once). For 4xx that are *not* 429 (e.g. 404), fail fast instead of
  retrying.
* **Concurrency:** the catalogue pages are independent — parallelize the
  page fetch with the same thread pool already used for `--detail`. For
  larger jobs, fan out over an asyncio + `httpx.AsyncClient` worker pool.
* **Proxies / IP pools:** when the target rate-limits per-IP, rotate a small
  set of residential/datacenter proxies per session and key your retry
  budget per-IP so a bad proxy doesn't poison the whole run.
* **Robustness:** add a JSON-Schema or `pandera` validation step on the
  parsed rows before the CSV write, so a silent layout change fails loudly
  instead of shipping a half-populated file.
* **Persistence:** for anything bigger than a few thousand rows, stream rows
  to SQLite/Parquet (or S3 + Parquet) instead of a single in-memory list, so
  a crash mid-run loses only the current page and the output is queryable.
* **Observability:** log per-page row counts and a run manifest (URLs,
  timestamps, retries) — it's the first thing you need when a downstream
  dashboard shows a gap.

## Files

* `scraper.py` — the scraper (this is the deliverable)
* `books.csv` — real output of a full run (1,000 rows)
* `requirements.txt` — `requests`, `beautifulsoup4`
* `README.md` — this file