mapillary-downloader 0.5.0__py3-none-any.whl → 0.5.1__py3-none-any.whl

mapillary_downloader/downloader.py

@@ -181,22 +181,22 @@ class MapillaryDownloader:
 
         start_time = time.time()
 
-        # Step 1: Build seen_ids from metadata file (streaming, only IDs)
-        logger.info("Building seen_ids from metadata...")
+        # Step 1: Check if API fetch is already complete
         reader = MetadataReader(self.metadata_file)
-        seen_ids = reader.get_all_ids()
         api_complete = reader.is_complete
-        logger.info(f"Found {len(seen_ids)} existing images in metadata")
+        if api_complete:
+            logger.info("API fetch already complete, will only download")
+        else:
+            logger.info("API fetch incomplete, will fetch and download in parallel")
 
-        # Step 2: Start worker pool (fork AFTER building seen_ids, BEFORE downloading)
-        pool = AdaptiveWorkerPool(
-            worker_process, min_workers=max(1, self.workers // 2), max_workers=self.workers, monitoring_interval=30
-        )
+        # Step 2: Start worker pool
+        # Since workers do both I/O (download) and CPU (WebP), need many more workers
+        # Cap at 128 for now - will build proper dynamic scaling on a new branch later
+        pool = AdaptiveWorkerPool(worker_process, min_workers=self.workers, max_workers=128, monitoring_interval=10)
         pool.start()
 
-        # Step 3: Download images from existing metadata while fetching new from API
+        # Step 3: Download images from metadata file while fetching new from API
         downloaded_count = 0
-        skipped = 0
         total_bytes = 0
         failed_count = 0
         submitted = 0
@@ -218,25 +218,18 @@ class MapillaryDownloader:
                         logger.info("API fetch thread: Starting...")
                         with open(self.metadata_file, "a") as meta_f:
                             for image in self.client.get_user_images(self.username, bbox=bbox):
-                                image_id = image["id"]
-
-                                # Skip if we already have this in our metadata file
-                                if image_id in seen_ids:
-                                    continue
-
-                                seen_ids.add(image_id)
                                 new_images_count[0] += 1
 
-                                # Save new metadata
+                                # Save metadata (don't dedupe here, let the tailer handle it)
                                 meta_f.write(json.dumps(image) + "\n")
                                 meta_f.flush()
 
                                 if new_images_count[0] % 1000 == 0:
-                                    logger.info(f"API: Fetched {new_images_count[0]} new images from API")
+                                    logger.info(f"API: Fetched {new_images_count[0]} images from API")
 
                             # Mark as complete
                             MetadataReader.mark_complete(self.metadata_file)
-                            logger.info(f"API fetch complete: {new_images_count[0]} new images")
+                            logger.info(f"API fetch complete: {new_images_count[0]} images")
                     finally:
                         api_fetch_complete.set()
 
@@ -438,14 +431,7 @@ class MapillaryDownloader:
         self._save_progress()
         elapsed = time.time() - start_time
 
-        # Count total images in metadata
-        total_images = len(seen_ids)
-        skipped = total_images - downloaded_count - failed_count
-
-        logger.info(
-            f"Complete! Total {total_images} images, downloaded {downloaded_count} ({format_size(total_bytes)}), "
-            f"skipped {skipped}, failed {failed_count}"
-        )
+        logger.info(f"Complete! Downloaded {downloaded_count} ({format_size(total_bytes)}), " f"failed {failed_count}")
         logger.info(f"Total time: {format_time(elapsed)}")
 
         # Tar sequence directories for efficient IA uploads

mapillary_downloader/metadata_reader.py

@@ -23,7 +23,47 @@ class MetadataReader:
             metadata_file: Path to metadata.jsonl or metadata.jsonl.gz
         """
         self.metadata_file = Path(metadata_file)
-        self.is_complete = False
+        self.is_complete = self._check_complete()
+
+    def _check_complete(self):
+        """Check if metadata file has completion marker.
+
+        Returns:
+            True if completion marker found, False otherwise
+        """
+        if not self.metadata_file.exists():
+            return False
+
+        # Check last few lines for completion marker (it should be at the end)
+        try:
+            if self.metadata_file.suffix == ".gz":
+                file_handle = gzip.open(self.metadata_file, "rt")
+            else:
+                file_handle = open(self.metadata_file)
+
+            with file_handle as f:
+                # Read last 10 lines to find completion marker
+                lines = []
+                for line in f:
+                    lines.append(line)
+                    if len(lines) > 10:
+                        lines.pop(0)
+
+                # Check if any of the last lines is the completion marker
+                for line in reversed(lines):
+                    line = line.strip()
+                    if not line:
+                        continue
+                    try:
+                        data = json.loads(line)
+                        if data.get("__complete__"):
+                            return True
+                    except json.JSONDecodeError:
+                        continue
+
+            return False
+        except Exception:
+            return False
 
     def iter_images(self, quality_field=None, downloaded_ids=None):
         """Stream images from metadata file with filtering.

mapillary_downloader/worker.py

@@ -17,6 +17,9 @@ def worker_process(work_queue, result_queue, worker_id):
         result_queue: Queue to push results to
         worker_id: Unique worker identifier
     """
+    # Create session once per worker (reuse HTTP connections)
+    session = requests.Session()
+
     while True:
         work_item = work_queue.get()
 
@@ -27,14 +30,17 @@ def worker_process(work_queue, result_queue, worker_id):
         # Unpack work item
         image_data, output_dir, quality, convert_webp, access_token = work_item
 
+        # Update session auth for this request
+        session.headers.update({"Authorization": f"OAuth {access_token}"})
+
         # Process the image
-        result = download_and_convert_image(image_data, output_dir, quality, convert_webp, access_token)
+        result = download_and_convert_image(image_data, output_dir, quality, convert_webp, session)
 
         # Push result back
         result_queue.put(result)
 
 
-def download_and_convert_image(image_data, output_dir, quality, convert_webp, access_token):
+def download_and_convert_image(image_data, output_dir, quality, convert_webp, session):
     """Download and optionally convert a single image.
 
     This function is designed to run in a worker process.
@@ -44,7 +50,7 @@ def download_and_convert_image(image_data, output_dir, quality, convert_webp, ac
         output_dir: Base output directory path
         quality: Quality level (256, 1024, 2048, original)
         convert_webp: Whether to convert to WebP
-        access_token: Mapillary API access token
+        session: requests.Session with auth already configured
 
     Returns:
         Tuple of (image_id, bytes_downloaded, success, error_msg)
@@ -78,11 +84,7 @@ def download_and_convert_image(image_data, output_dir, quality, convert_webp, ac
             jpg_path = img_dir / f"{image_id}.jpg"
             final_path = jpg_path
 
-        # Download image
-        # No retries for CDN images - they're cheap, just skip failures and move on
-        session = requests.Session()
-        session.headers.update({"Authorization": f"OAuth {access_token}"})
-
+        # Download image (using session passed from worker)
         bytes_downloaded = 0
 
         try:

mapillary_downloader/worker_pool.py

@@ -17,7 +17,7 @@ class AdaptiveWorkerPool:
     - If throughput plateauing/decreasing: reduce workers
     """
 
-    def __init__(self, worker_func, min_workers=4, max_workers=16, monitoring_interval=30):
+    def __init__(self, worker_func, min_workers=4, max_workers=16, monitoring_interval=10):
         """Initialize adaptive worker pool.
 
         Args:
@@ -37,10 +37,11 @@ class AdaptiveWorkerPool:
 
         # Worker management
         self.workers = []
-        self.current_workers = min_workers
+        self.current_workers = min_workers  # Start small and ramp up
 
         # Throughput monitoring
         self.throughput_history = deque(maxlen=5)  # Last 5 measurements
+        self.worker_count_history = deque(maxlen=5)  # Track worker counts at each measurement
         self.last_processed = 0
         self.last_check_time = time.time()
 
@@ -86,36 +87,103 @@ class AdaptiveWorkerPool:
         elapsed = now - self.last_check_time
 
         if elapsed < self.monitoring_interval:
+            logger.debug(f"Throughput check skipped (elapsed {elapsed:.1f}s < {self.monitoring_interval}s)")
             return
 
         # Calculate current throughput (items/sec)
         items_since_check = total_processed - self.last_processed
         throughput = items_since_check / elapsed
 
+        current_workers = len(self.workers)
         self.throughput_history.append(throughput)
+        self.worker_count_history.append(current_workers)
         self.last_processed = total_processed
         self.last_check_time = now
 
-        # Need at least 3 measurements to detect trends
-        if len(self.throughput_history) < 3:
+        logger.info(
+            f"Throughput: {throughput:.1f} items/s (workers: {current_workers}/{self.max_workers}, "
+            f"history: {len(self.throughput_history)} measurements)"
+        )
+
+        # Need at least 2 measurements to calculate gain per worker
+        if len(self.throughput_history) < 2:
+            # First measurement - add 20% more workers
+            if current_workers < self.max_workers:
+                workers_to_add = max(1, int(current_workers * 0.2))
+                for i in range(workers_to_add):
+                    if len(self.workers) < self.max_workers:
+                        new_worker_id = len(self.workers)
+                        self._add_worker(new_worker_id)
+                        self.current_workers += 1
+                logger.info(
+                    f"Ramping up: added {workers_to_add} workers (now {self.current_workers}/{self.max_workers})"
+                )
             return
 
-        # Check if throughput is increasing
-        recent_avg = sum(list(self.throughput_history)[-2:]) / 2
-        older_avg = sum(list(self.throughput_history)[-4:-2]) / 2
-
-        if recent_avg > older_avg * 1.1 and len(self.workers) < self.max_workers:
-            # Throughput increasing by >10%, add workers
-            new_worker_id = len(self.workers)
-            self._add_worker(new_worker_id)
-            self.current_workers += 1
-            logger.info(f"Throughput increasing ({throughput:.1f} items/s), added worker (now {self.current_workers})")
-
-        elif recent_avg < older_avg * 0.9 and len(self.workers) > self.min_workers:
-            # Throughput decreasing by >10%, remove worker
-            # (workers will exit naturally when they finish current work)
-            self.current_workers = max(self.min_workers, self.current_workers - 1)
-            logger.info(f"Throughput plateauing ({throughput:.1f} items/s), reducing to {self.current_workers} workers")
+        # Calculate throughput gain per worker added
+        current_throughput = self.throughput_history[-1]
+        previous_throughput = self.throughput_history[-2]
+        previous_workers = self.worker_count_history[-2]
+
+        throughput_gain = current_throughput - previous_throughput
+        workers_added = current_workers - previous_workers
+
+        logger.debug(
+            f"Trend: {previous_throughput:.1f} items/s @ {previous_workers} workers → "
+            f"{current_throughput:.1f} items/s @ {current_workers} workers "
+            f"(gain: {throughput_gain:.1f}, added: {workers_added})"
+        )
+
+        # If throughput decreased significantly, stop adding workers
+        if current_throughput < previous_throughput * 0.95:
+            logger.info(
+                f"Throughput decreasing ({current_throughput:.1f} vs {previous_throughput:.1f} items/s), "
+                f"stopping at {current_workers} workers"
+            )
+        # If throughput is still increasing or stable, add more workers
+        elif current_throughput >= previous_throughput * 0.95 and current_workers < self.max_workers:
+            if workers_added > 0 and throughput_gain > 0:
+                # Calculate gain per worker
+                gain_per_worker = throughput_gain / workers_added
+                logger.debug(f"Gain per worker: {gain_per_worker:.2f} items/s")
+
+                # Estimate how many more workers we could benefit from
+                # Assume diminishing returns, so be conservative
+                if gain_per_worker > 0.5:
+                    # Good gain per worker - add more aggressively
+                    workers_to_add = max(1, int(current_workers * 0.3))
+                elif gain_per_worker > 0.2:
+                    # Moderate gain - add moderately
+                    workers_to_add = max(1, int(current_workers * 0.2))
+                else:
+                    # Small gain - add conservatively
+                    workers_to_add = max(1, int(current_workers * 0.1))
+
+                added = 0
+                for i in range(workers_to_add):
+                    if len(self.workers) < self.max_workers:
+                        new_worker_id = len(self.workers)
+                        self._add_worker(new_worker_id)
+                        self.current_workers += 1
+                        added += 1
+
+                logger.info(
+                    f"Throughput increasing (gain: {gain_per_worker:.2f} items/s per worker), "
+                    f"added {added} workers (now {self.current_workers}/{self.max_workers})"
+                )
+            else:
+                # Fallback to 20% if we can't calculate gain per worker
+                workers_to_add = max(1, int(current_workers * 0.2))
+                added = 0
+                for i in range(workers_to_add):
+                    if len(self.workers) < self.max_workers:
+                        new_worker_id = len(self.workers)
+                        self._add_worker(new_worker_id)
+                        self.current_workers += 1
+                        added += 1
+                logger.info(f"Ramping up: added {added} workers (now {self.current_workers}/{self.max_workers})")
+        else:
+            logger.info(f"At optimal worker count: {current_workers} workers, {current_throughput:.1f} items/s")
 
     def shutdown(self, timeout=30):
         """Shutdown the worker pool gracefully."""

{mapillary_downloader-0.5.0.dist-info → mapillary_downloader-0.5.1.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mapillary_downloader
-Version: 0.5.0
+Version: 0.5.1
 Summary: Download your Mapillary data before it's gone
 Author-email: Gareth Davidson <gaz@bitplane.net>
 Requires-Python: >=3.10

{mapillary_downloader-0.5.0.dist-info → mapillary_downloader-0.5.1.dist-info}/RECORD

@@ -1,19 +1,19 @@
 mapillary_downloader/__init__.py,sha256=KEjiBRghXDeA7E15RJeLBfQm-yNJkowZarL59QOh_1w,120
 mapillary_downloader/__main__.py,sha256=Kjfx2woMyCvAxYAdqvtXtYJknCMviV_K2PSo0cDc8Hg,4320
 mapillary_downloader/client.py,sha256=a5n43FLHP45EHodEjl0ieziBK-b6Ey-rZJwYB6EFhNI,4743
-mapillary_downloader/downloader.py,sha256=F36AtB0Ro_EXR78EDOqH248llV7fGVeR4j9nZf0q7qg,19988
+mapillary_downloader/downloader.py,sha256=v0vLovW80DMpEzQIb_tdETF1HzO2GEMbbVzD93aKnnQ,19561
 mapillary_downloader/exif_writer.py,sha256=K_441EG1siWyNMmFGZSfnORUCjBThkeg4JFtbg9AOsA,5120
 mapillary_downloader/ia_check.py,sha256=L2MEbG_KmlAd5NLmo2HQkO8HWvRN0brE5wXXoyNMbq8,1100
 mapillary_downloader/ia_meta.py,sha256=78rcybHIPnQDsF02KGj6RYmDXzYzrU8sdVx4Q9Y0sfI,6266
 mapillary_downloader/logging_config.py,sha256=Z-wNq34nt7aIhJWdeKc1feTY46P9-Or7HtiX7eUFjEI,2324
-mapillary_downloader/metadata_reader.py,sha256=-4BmtLVI9sldZU0LlqMc-bporiYNpk6-F2RKKMvzLu4,3560
+mapillary_downloader/metadata_reader.py,sha256=Re-HN0Vfc7Hs1eOut7uOoW7jWJ2PIbKoNzC7Ak3ah5o,4933
 mapillary_downloader/tar_sequences.py,sha256=mqs5p3N7osV_bxTkw6i34GVmxCBBEbIiKKxeh-fWNdU,4430
 mapillary_downloader/utils.py,sha256=yzVgS1mwsklDAqrimaFafgTTXtRYQUbKP98Xgh9d2KA,1174
 mapillary_downloader/webp_converter.py,sha256=vYLLQxDmdnqRz0nm7wXwRUd4x9mQZNah-DrncpA8sNs,1901
-mapillary_downloader/worker.py,sha256=RMZO8N67Kl-bhHC1qUdZg6Sx8k6RYbPRhyuLyOjr29o,4450
-mapillary_downloader/worker_pool.py,sha256=QFYIbqkgamOtB-iRyZp5kN6jdZuYw93izls61ayVIZ8,4771
-mapillary_downloader-0.5.0.dist-info/entry_points.txt,sha256=PdYtxOXHMJrUhmiPO4G-F98VuhUI4MN9D_T4KPrVZ5w,75
-mapillary_downloader-0.5.0.dist-info/licenses/LICENSE.md,sha256=7_BIuQ-veOrsF-WarH8kTkm0-xrCLvJ1PFE1C4Ebs64,146
-mapillary_downloader-0.5.0.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
-mapillary_downloader-0.5.0.dist-info/METADATA,sha256=A0AhsIjGV9FBf5vz28hSC2jugcRqz5A8gsZwMGGEw2A,4982
-mapillary_downloader-0.5.0.dist-info/RECORD,,
+mapillary_downloader/worker.py,sha256=syxsE2pPX_9EXzSGEGeUaeLFqmqZjTma-rB7S2zpYac,4511
+mapillary_downloader/worker_pool.py,sha256=ctFl40UgFUjpLL_e6Mw5h7YNMfKwXulRhaX18r9sIkE,8257
+mapillary_downloader-0.5.1.dist-info/entry_points.txt,sha256=PdYtxOXHMJrUhmiPO4G-F98VuhUI4MN9D_T4KPrVZ5w,75
+mapillary_downloader-0.5.1.dist-info/licenses/LICENSE.md,sha256=7_BIuQ-veOrsF-WarH8kTkm0-xrCLvJ1PFE1C4Ebs64,146
+mapillary_downloader-0.5.1.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
+mapillary_downloader-0.5.1.dist-info/METADATA,sha256=1vM80Kq2WHUzfSt7lIx91hb0fycHosxpll-xnCda6JU,4982
+mapillary_downloader-0.5.1.dist-info/RECORD,,