Files
scaev/docs/ARCHITECTURE.md
2025-12-07 07:09:16 +01:00

532 lines
30 KiB
Markdown

# Scaev - Architecture & Data Flow
## System Overview
The scraper follows a **3-phase hierarchical crawling pattern** to extract auction and lot data from Troostwijk Auctions website.
## Architecture Diagram
```mariadb
TROOSTWIJK SCRAPER
PHASE 1: COLLECT AUCTION URLs
Listing Page Extract /a/
/auctions? auction URLs
page=1..N
[ List of Auction URLs ]
PHASE 2: EXTRACT LOT URLs FROM AUCTIONS
Auction Page Parse
/a/... __NEXT_DATA__
JSON
Save Auction Extract /l/
Metadata lot URLs
to DB
[ List of Lot URLs ]
PHASE 3: SCRAPE LOT DETAILS + API ENRICHMENT
Lot Page Parse
/l/... __NEXT_DATA__
JSON
GraphQL API Bid History Save Images
(Bidding + REST API URLs to DB
Enrichment) (per lot)
[Optional Download
Concurrent per Lot]
Save to DB:
- Lot data
- Bid data
- Enrichment
```
## Database Schema
```mariadb
CACHE TABLE (HTML Storage with Compression)
cache
url (TEXT, PRIMARY KEY)
content (BLOB) -- Compressed HTML (zlib) │
timestamp (REAL)
status_code (INTEGER)
compressed (INTEGER) -- 1=compressed, 0=plain │
AUCTIONS TABLE
auctions
auction_id (TEXT, PRIMARY KEY) -- e.g. "A7-39813" │
url (TEXT, UNIQUE)
title (TEXT)
location (TEXT) -- e.g. "Cluj-Napoca, RO" │
lots_count (INTEGER)
first_lot_closing_time (TEXT)
scraped_at (TEXT)
LOTS TABLE (Core + Enriched Intelligence)
lots
lot_id (TEXT, PRIMARY KEY) -- e.g. "A1-28505-5" │
auction_id (TEXT) -- FK to auctions │
url (TEXT, UNIQUE)
title (TEXT)
BIDDING DATA (GraphQL API)
current_bid (TEXT) -- Current bid amount │
starting_bid (TEXT) -- Initial/opening bid │
minimum_bid (TEXT) -- Next minimum bid │
bid_count (INTEGER) -- Number of bids │
bid_increment (REAL) -- Bid step size │
closing_time (TEXT) -- Lot end date │
status (TEXT) -- Minimum bid status │
BID INTELLIGENCE (Calculated from bid_history)
first_bid_time (TEXT) -- First bid timestamp │
last_bid_time (TEXT) -- Latest bid timestamp │
bid_velocity (REAL) -- Bids per hour │
VALUATION & ATTRIBUTES (from __NEXT_DATA__)
brand (TEXT) -- Brand from attributes │
model (TEXT) -- Model from attributes │
manufacturer (TEXT) -- Manufacturer name │
year_manufactured (INTEGER) -- Year extracted │
condition_score (REAL) -- 0-10 condition rating │
condition_description (TEXT) -- Condition text │
serial_number (TEXT) -- Serial/VIN number │
damage_description (TEXT) -- Damage notes │
attributes_json (TEXT) -- Full attributes JSON │
LEGACY/OTHER
viewing_time (TEXT) -- Viewing schedule │
pickup_date (TEXT) -- Pickup schedule │
location (TEXT) -- e.g. "Dongen, NL" │
description (TEXT) -- Lot description │
category (TEXT) -- Lot category │
sale_id (INTEGER) -- Legacy field │
type (TEXT) -- Legacy field │
year (INTEGER) -- Legacy field │
currency (TEXT) -- Currency code │
closing_notified (INTEGER) -- Notification flag │
scraped_at (TEXT) -- Scrape timestamp │
FOREIGN KEY (auction_id) auctions(auction_id)
IMAGES TABLE (Image URLs & Download Status)
images THIS TABLE HOLDS IMAGE LINKS
id (INTEGER, PRIMARY KEY AUTOINCREMENT)
lot_id (TEXT) -- FK to lots │
url (TEXT) -- Image URL │
local_path (TEXT) -- Path after download │
downloaded (INTEGER) -- 0=pending, 1=downloaded │
FOREIGN KEY (lot_id) lots(lot_id)
UNIQUE INDEX idx_unique_lot_url ON (lot_id, url)
BID_HISTORY TABLE (Complete Bid Tracking for Intelligence)
bid_history REST API: /bidding-history
id (INTEGER, PRIMARY KEY AUTOINCREMENT)
lot_id (TEXT) -- FK to lots │
bid_amount (REAL) -- Bid in EUR │
bid_time (TEXT) -- ISO 8601 timestamp │
is_autobid (INTEGER) -- 0=manual, 1=autobid │
bidder_id (TEXT) -- Anonymized bidder UUID │
bidder_number (INTEGER) -- Bidder display number │
created_at (TEXT) -- Record creation timestamp │
FOREIGN KEY (lot_id) lots(lot_id)
INDEX idx_bid_history_lot ON (lot_id)
INDEX idx_bid_history_time ON (bid_time)
```
## Sequence Diagram
```
User Scraper Playwright Cache DB Data Tables
│ │ │ │ │
│ Run │ │ │ │
├──────────────▶│ │ │ │
│ │ │ │ │
│ │ Phase 1: Listing Pages │ │
│ ├───────────────▶│ │ │
│ │ goto() │ │ │
│ │◀───────────────┤ │ │
│ │ HTML │ │ │
│ ├───────────────────────────────▶│ │
│ │ compress & cache │ │
│ │ │ │ │
│ │ Phase 2: Auction Pages │ │
│ ├───────────────▶│ │ │
│ │◀───────────────┤ │ │
│ │ HTML │ │ │
│ │ │ │ │
│ │ Parse __NEXT_DATA__ JSON │ │
│ │────────────────────────────────────────────────▶│
│ │ │ │ INSERT auctions
│ │ │ │ │
│ │ Phase 3: Lot Pages │ │
│ ├───────────────▶│ │ │
│ │◀───────────────┤ │ │
│ │ HTML │ │ │
│ │ │ │ │
│ │ Parse __NEXT_DATA__ JSON │ │
│ │────────────────────────────────────────────────▶│
│ │ │ │ INSERT lots │
│ │────────────────────────────────────────────────▶│
│ │ │ │ INSERT images│
│ │ │ │ │
│ │ Export to CSV/JSON │ │
│ │◀────────────────────────────────────────────────┤
│ │ Query all data │ │
│◀──────────────┤ │ │ │
│ Results │ │ │ │
```
## Data Flow Details
### 1. **Page Retrieval & Caching**
```
Request URL
├──▶ Check cache DB (with timestamp validation)
│ │
│ ├─[HIT]──▶ Decompress (if compressed=1)
│ │ └──▶ Return HTML
│ │
│ └─[MISS]─▶ Fetch via Playwright
│ │
│ ├──▶ Compress HTML (zlib level 9)
│ │ ~70-90% size reduction
│ │
│ └──▶ Store in cache DB (compressed=1)
└──▶ Return HTML for parsing
```
### 2. **JSON Parsing Strategy**
```
HTML Content
└──▶ Extract <script id="__NEXT_DATA__">
├──▶ Parse JSON
│ │
│ ├─[has pageProps.lot]──▶ Individual LOT
│ │ └──▶ Extract: title, bid, location, images, etc.
│ │
│ └─[has pageProps.auction]──▶ AUCTION
│ │
│ ├─[has lots[] array]──▶ Auction with lots
│ │ └──▶ Extract: title, location, lots_count
│ │
│ └─[no lots[] array]──▶ Old format lot
│ └──▶ Parse as lot
└──▶ Fallback to HTML regex parsing (if JSON fails)
```
### 3. **API Enrichment Flow**
```
Lot Page Scraped (__NEXT_DATA__ parsed)
├──▶ Extract lot UUID from JSON
├──▶ GraphQL API Call (fetch_lot_bidding_data)
│ └──▶ Returns: current_bid, starting_bid, minimum_bid,
│ bid_count, closing_time, status, bid_increment
├──▶ [If bid_count > 0] REST API Call (fetch_bid_history)
│ │
│ ├──▶ Fetch all bid pages (paginated)
│ │
│ └──▶ Returns: Complete bid history with timestamps,
│ bidder_ids, autobid flags, amounts
│ │
│ ├──▶ INSERT INTO bid_history (multiple records)
│ │
│ └──▶ Calculate bid intelligence:
│ - first_bid_time (earliest timestamp)
│ - last_bid_time (latest timestamp)
│ - bid_velocity (bids per hour)
├──▶ Extract enrichment from __NEXT_DATA__:
│ - Brand, model, manufacturer (from attributes)
│ - Year (regex from title/attributes)
│ - Condition (map to 0-10 score)
│ - Serial number, damage description
└──▶ INSERT/UPDATE lots table with all data
```
### 4. **Image Handling (Concurrent per Lot)**
```
Lot Page Parsed
├──▶ Extract images[] from JSON
│ │
│ └──▶ INSERT OR IGNORE INTO images (lot_id, url, downloaded=0)
│ └──▶ Unique constraint prevents duplicates
└──▶ [If DOWNLOAD_IMAGES=True]
├──▶ Create concurrent download tasks (asyncio.gather)
│ │
│ ├──▶ All images for lot download in parallel
│ │ (No rate limiting between images in same lot)
│ │
│ ├──▶ Save to: /images/{lot_id}/001.jpg
│ │
│ └──▶ UPDATE images SET local_path=?, downloaded=1
└──▶ Rate limit only between lots (0.5s)
(Not between images within a lot)
```
## Key Configuration
| Setting | Value | Purpose |
|----------------------|-----------------------------------|----------------------------------|
| `CACHE_DB` | `/mnt/okcomputer/output/cache.db` | SQLite database path |
| `IMAGES_DIR` | `/mnt/okcomputer/output/images` | Downloaded images storage |
| `RATE_LIMIT_SECONDS` | `0.5` | Delay between requests |
| `DOWNLOAD_IMAGES` | `False` | Toggle image downloading |
| `MAX_PAGES` | `50` | Number of listing pages to crawl |
## Output Files
```
/mnt/okcomputer/output/
├── cache.db # SQLite database (compressed HTML + data)
├── auctions_{timestamp}.json # Exported auctions
├── auctions_{timestamp}.csv # Exported auctions
├── lots_{timestamp}.json # Exported lots
├── lots_{timestamp}.csv # Exported lots
└── images/ # Downloaded images (if enabled)
├── A1-28505-5/
│ ├── 001.jpg
│ └── 002.jpg
└── A1-28505-6/
└── 001.jpg
```
## Extension Points for Integration
### 1. **Downstream Processing Pipeline**
```sqlite
-- Query lots without downloaded images
SELECT lot_id, url FROM images WHERE downloaded = 0;
-- Process images: OCR, classification, etc.
-- Update status when complete
UPDATE images SET downloaded = 1, local_path = ? WHERE id = ?;
```
### 2. **Real-time Monitoring**
```sqlite
-- Check for new lots every N minutes
SELECT COUNT(*) FROM lots WHERE scraped_at > datetime('now', '-1 hour');
-- Monitor bid changes
SELECT lot_id, current_bid, bid_count FROM lots WHERE bid_count > 0;
```
### 3. **Analytics & Reporting**
```sqlite
-- Top locations
SELECT location, COUNT(*) as lots_count FROM lots GROUP BY location;
-- Auction statistics
SELECT
a.auction_id,
a.title,
COUNT(l.lot_id) as actual_lots,
SUM(CASE WHEN l.bid_count > 0 THEN 1 ELSE 0 END) as lots_with_bids
FROM auctions a
LEFT JOIN lots l ON a.auction_id = l.auction_id
GROUP BY a.auction_id
```
### 4. **Image Processing Integration**
```sqlite
-- Get all images for a lot
SELECT url, local_path FROM images WHERE lot_id = 'A1-28505-5';
-- Batch process unprocessed images
SELECT i.id, i.lot_id, i.local_path, l.title, l.category
FROM images i
JOIN lots l ON i.lot_id = l.lot_id
WHERE i.downloaded = 1 AND i.local_path IS NOT NULL;
```
## Performance Characteristics
- **Compression**: ~70-90% HTML size reduction (1GB → ~100-300MB)
- **Rate Limiting**: Exactly 0.5s between requests (respectful scraping)
- **Caching**: 24-hour default cache validity (configurable)
- **Throughput**: ~7,200 pages/hour (with 0.5s rate limit)
- **Scalability**: SQLite handles millions of rows efficiently
## Error Handling
- **Network failures**: Cached as status_code=500, retry after cache expiry
- **Parse failures**: Falls back to HTML regex patterns
- **Compression errors**: Auto-detects and handles uncompressed legacy data
- **Missing fields**: Defaults to "No bids", empty string, or 0
## Rate Limiting & Ethics
- **REQUIRED**: 0.5 second delay between page requests (not between images)
- **Respects cache**: Avoids unnecessary re-fetching
- **User-Agent**: Identifies as standard browser
- **No parallelization**: Single-threaded sequential crawling for pages
- **Image downloads**: Concurrent within each lot (16x speedup)
---
## API Integration Architecture
### GraphQL API
**Endpoint:** `https://storefront.tbauctions.com/storefront/graphql`
**Purpose:** Real-time bidding data and lot enrichment
**Key Query:**
```graphql
query LotBiddingData($lotDisplayId: String!, $locale: String!, $platform: Platform!) {
lotDetails(displayId: $lotDisplayId, locale: $locale, platform: $platform) {
lot {
currentBidAmount { cents currency }
initialAmount { cents currency }
nextMinimalBid { cents currency }
nextBidStepInCents
bidsCount
followersCount # Available - Watch count
startDate
endDate
minimumBidAmountMet
biddingStatus
condition
location { city countryCode }
categoryInformation { name path }
attributes { name value }
}
estimatedFullPrice { # Available - Estimated value
min { cents currency }
max { cents currency }
}
}
}
```
**Currently Captured:**
- ✅ Current bid, starting bid, minimum bid
- ✅ Bid count and bid increment
- ✅ Closing time and status
- ✅ Brand, model, manufacturer (from attributes)
**Available but Not Yet Captured:**
- ⚠️ `followersCount` - Watch count for popularity analysis
- ⚠️ `estimatedFullPrice` - Min/max estimated values
- ⚠️ `biddingStatus` - More detailed status enum
- ⚠️ `condition` - Direct condition field
- ⚠️ `location` - City, country details
- ⚠️ `categoryInformation` - Structured category
### REST API - Bid History
**Endpoint:** `https://shared-api.tbauctions.com/bidmanagement/lots/{lot_uuid}/bidding-history`
**Purpose:** Complete bid history for intelligence analysis
**Parameters:**
- `pageNumber` (starts at 1)
- `pageSize` (default: 100)
**Response Example:**
```json
{
"results": [
{
"buyerId": "uuid", // Anonymized bidder ID
"buyerNumber": 4, // Display number
"currentBid": {
"cents": 370000,
"currency": "EUR"
},
"autoBid": false, // Is autobid
"negotiated": false, // Was negotiated
"createdAt": "2025-12-05T04:53:56.763033Z"
}
],
"hasNext": true,
"pageNumber": 1
}
```
**Captured Data:**
- ✅ Bid amount, timestamp, bidder ID
- ✅ Autobid flag
- ⚠️ `negotiated` - Not yet captured
**Calculated Intelligence:**
- ✅ First bid time
- ✅ Last bid time
- ✅ Bid velocity (bids per hour)
### API Integration Points
**Files:**
- `src/graphql_client.py` - GraphQL queries and parsing
- `src/bid_history_client.py` - REST API pagination and parsing
- `src/scraper.py` - Integration during lot scraping
**Flow:**
1. Lot page scraped → Extract lot UUID from `__NEXT_DATA__`
2. Call GraphQL API → Get bidding data
3. If bid_count > 0 → Call REST API → Get complete bid history
4. Calculate bid intelligence metrics
5. Save to database
**Rate Limiting:**
- API calls happen during lot scraping phase
- Overall 0.5s rate limit applies to page requests
- API calls are part of lot processing (not separately limited)
See `API_INTELLIGENCE_FINDINGS.md` for detailed field analysis and roadmap.