# Parallel Scrapers with Country Mapping Fix

## Problem

The scheduler runs scrapers sequentially — one language at a time. With 19,996 unscraped churches queued across 10 language scrapers, a full cycle takes days. The English scraper alone runs 30+ hours. Additionally, 1,414 churches in unmapped countries (BE, CH, IN, etc.) fall through to the generic scraper instead of being handled by appropriate language scrapers.

## Changes

### 1. Country Mapping Additions (scraper-service.ts)

Add to `COUNTRY_SCRAPER_MAP`:
- **English**: IN, SG, MY, KE, JM, TT, GH, NG, ZA, TZ, UG
- **French**: BE, LU
- **German**: CH, SI
- **Italian**: HR, RO

### 2. Parallel Pipeline Groups (scheduler.ts)

Replace sequential `PIPELINE_PHASES` array with grouped phases:

| Group | Phases | Concurrency |
|-------|--------|-------------|
| 1 | osm-import, gcatholic-import | Sequential (shared data) |
| 2 | english, french, german | Parallel (3) |
| 3 | polish, spanish, italian | Parallel (3) |
| 4 | portuguese, czech, dutch | Parallel (3) |
| 5 | hungarian, generic | Parallel (2) |

Scheduler starts all jobs in a group simultaneously, waits for all to finish, then advances to the next group.

### 3. Generic Scraper Deprioritized

- Moved to last group
- Pre-check query: skip if no unscraped churches in generic queue (avoids wasteful re-scrapes)

### 4. Resource Changes

- Scheduler container memory limit: 4GB → 10GB (3 concurrent Playwright/Chromium processes)
- No new Docker containers or compose changes needed — existing child process spawning approach is kept

## Approach

Approach B: parallel child processes inside the scheduler container. No Docker-in-Docker. The scheduler already spawns `npx tsx` processes — we just allow multiple to run concurrently instead of waiting for each to finish before starting the next.