write positions in 10-minute batches

src/ptscrapper/main.py

@@ -3,6 +3,7 @@
 import asyncio
 import logging
 
+import pyarrow as pa
 import uvloop
 
 from ptscrapper.scheduler import start_scheduler
@@ -14,6 +15,8 @@ try:
 except ImportError:
     pass
 
+pa.set_cpu_count(1)
+
 
 def main():
     # 1) Configure the root logger

src/ptscrapper/scheduler.py

@@ -32,7 +32,7 @@ from .storage import (
     push_to_nas,
     write_course_geometry,
     write_course_posts,
-    write_positions,
+    write_positions_rb,
     write_posts_lines,
     write_stops,
 )
@@ -120,22 +120,9 @@ async def positions_job():
         server_ts = datetime.fromisoformat(server_ts_str).replace(
             tzinfo=ZoneInfo("Europe/Warsaw")
         )
-
-        # 3) build Arrow table enrich each dict with both timestamps
-        enriched = []
-        for p in positions:
-            enriched.append(
-                {
-                    **p,
-                    "server_ts": server_ts,
-                    "fetched_ts": fetch_ts,
-                }
-            )
-        table = pa.Table.from_pylist(enriched)
-
-        # 4) pass both timestamps through
-        write_positions(table, server_ts)
-        logger.info("wrote positions shard for %s", server_ts)
+        # 3) hand off raw list to our super-light RecordBatchBuilder
+        write_positions_rb(positions, server_ts, fetch_ts)
+        logger.info("appended %d rows into 10 min shard %s", len(positions), server_ts)
     except Exception:
         # record error details
         logger.exception("job=positions_job")

src/ptscrapper/storage.py

@@ -2,6 +2,7 @@
 
 import glob
 import os
+import threading
 from datetime import date, datetime
 
 import geoarrow.pyarrow as ga
@@ -9,6 +10,7 @@ import polyline
 import pyarrow as pa
 import pyarrow.dataset as ds
 import pyarrow.parquet as pq
+from pyarrow import lib as _lib
 from rclone_python import rclone
 
 from .config import HOST_ID  # unique identifier for this host/container
@@ -18,29 +20,112 @@ from .config import (  # base path on the remote, e.g. "pt-scraper-data"
     RCLONE_REMOTE_PATH,
 )
 
+# a fixed schema for every positions batch
+POSITIONS_SCHEMA = pa.schema(
+    [
+        ("v", pa.int64()),
+        ("c", pa.int64()),
+        ("x", pa.float64()),
+        ("y", pa.float64()),
+        ("l", pa.string()),
+        ("t", pa.string()),
+        ("s", pa.string()),
+        ("d", pa.string()),
+        ("e", pa.int64()),
+        ("server_ts", pa.timestamp("ns", tz="Europe/Warsaw")),
+        ("fetched_ts", pa.timestamp("ns", tz="UTC")),
+    ]
+)
+
+# module-level rolling writer state
+_positions_lock = threading.Lock()
+_positions_writer = None  # pq.ParquetWriter
+_positions_bucket = None  # datetime start of current 10-min
+_positions_path = None  # path to current .parquet file
+
 
 def _ensure_dir(path: str):
     """Make sure the directory exists."""
     os.makedirs(path, exist_ok=True)
 
 
-def write_positions(
-    table: pa.Table,
-    server_ts: datetime,
-) -> str:
+def _rollover_writer(server_ts):
     """
-    Write out the raw positions (including x,y, server_ts, fetched_ts).
-    Geometry will be added later during daily compaction.
+    Close the old bucket (if any), push it via rclone,
+    then open a new ParquetWriter for the 10-min bucket of `server_ts`.
     """
+    global _positions_writer, _positions_bucket, _positions_path
 
-    subdir = server_ts.strftime("positions/%Y/%m/%d/%H")
+    # floor to ten-minute
+    bucket = server_ts.replace(
+        minute=(server_ts.minute // 10) * 10, second=0, microsecond=0
+    )
+
+    if bucket != _positions_bucket:
+        #  -- close previous shard
+        if _positions_writer is not None:
+            _positions_writer.close()
+
+        #  -- open new writer
+        subdir = bucket.strftime("positions/%Y/%m/%d/%H")
         local_dir = os.path.join(LOCAL_DATA_DIR, subdir)
         _ensure_dir(local_dir)
 
-    filename = f"{server_ts.strftime('%M%S')}_{HOST_ID}.parquet"
-    local_path = os.path.join(local_dir, filename)
-    pq.write_table(table, local_path, compression="snappy")
-    return local_path
+        fn = f"{bucket.strftime('%M%S')}_{HOST_ID}.parquet"
+        _positions_path = os.path.join(local_dir, fn)
+
+        _positions_writer = pq.ParquetWriter(
+            _positions_path,
+            schema=POSITIONS_SCHEMA,
+            compression="none",  # or "none" if you prefer to compress only at nightly
+            use_dictionary=False,
+            write_statistics=False,
+        )
+        _positions_bucket = bucket
+
+
+def write_positions_rb(positions: list[dict], server_ts, fetched_ts) -> None:
+    """
+    Append the incoming `positions` (list of dicts) into the
+    current 10-minute Parquet shard via a RecordBatchBuilder.
+    """
+    global _positions_writer
+    with _positions_lock:
+        _rollover_writer(server_ts)
+
+        # 1) prepare one builder per column in the same order as POSITIONS_SCHEMA
+        builders = [
+            _lib.Int64Builder(),  # v
+            _lib.Int64Builder(),  # c
+            _lib.Float64Builder(),  # x
+            _lib.Float64Builder(),  # y
+            _lib.StringBuilder(),  # l
+            _lib.StringBuilder(),  # t
+            _lib.StringBuilder(),  # s
+            _lib.StringBuilder(),  # d
+            _lib.Int64Builder(),  # e
+            _lib.TimestampBuilder("ns", tz="Europe/Warsaw"),  # server_ts
+            _lib.TimestampBuilder("ns", tz="UTC"),  # fetched_ts
+        ]
+
+        # 2) append into each builder
+        for p in positions:
+            builders[0].append(p["v"])
+            builders[1].append(p["c"])
+            builders[2].append(p["x"])
+            builders[3].append(p["y"])
+            builders[4].append(p["l"])
+            builders[5].append(p["t"])
+            builders[6].append(p["s"])
+            builders[7].append(p["d"])
+            builders[8].append(p["e"])
+            builders[9].append(server_ts)
+            builders[10].append(fetched_ts)
+
+        # 3) finish → arrays, make a RecordBatch, write it
+        arrays = [b.finish() for b in builders]
+        batch = pa.RecordBatch.from_arrays(arrays, schema=POSITIONS_SCHEMA)
+        _positions_writer.write_batch(batch)
 
 
 def write_course_posts(table: pa.Table, course_id: int, ts: datetime) -> str: