Refine init DB script

This commit is contained in:
Marcel Schwarz 2020-12-19 00:29:32 +01:00
parent 9f8076286e
commit 8b0a107c89

View File

@ -1,14 +1,18 @@
import csv
import json
import logging
import os
import os.path
import shutil
import sqlite3
import time
import urllib.request
from concurrent.futures import ThreadPoolExecutor
from dataclasses import dataclass
from datetime import datetime
import requests
WORKING_DIR = os.getcwd()
TMP_DIR = os.path.join(WORKING_DIR, "tmp")
@ -54,10 +58,11 @@ def get_online_files_list(subdir_filter=None, file_extension_filter=None):
class BikeDatabase:
def __init__(self):
self.conn = sqlite3.connect(os.path.join(WORKING_DIR, "bike-data.db"), check_same_thread=False, timeout=60)
self.cursor = self.conn.cursor()
LOG.info("Created new database connection")
self.conn = sqlite3.connect(os.path.join(WORKING_DIR, "bike-data.db"), check_same_thread=False, timeout=300)
def init_database(self):
LOG.info("Try to create tables")
self.conn.execute("""CREATE TABLE IF NOT EXISTS usage_stats(
rental_id INTEGER PRIMARY KEY,
duration INTEGER,
@ -70,6 +75,30 @@ class BikeDatabase:
start_station_name TEXT
)""")
self.conn.execute("CREATE TABLE IF NOT EXISTS read_files(file_path TEXT, etag TEXT PRIMARY KEY)")
self.conn.execute("""CREATE TABLE IF NOT EXISTS bike_points(
id TEXT PRIMARY KEY,
common_name TEXT,
lat REAL,
lon REAL,
id_num INTEGER AS (CAST(SUBSTR(id, 12) as INTEGER)) STORED
)""")
self.conn.commit()
LOG.info("Tables created")
def create_indexes(self):
LOG.info("Try to create indexes")
self.conn.execute("""CREATE INDEX IF NOT EXISTS idx_date_of_start_date
ON usage_stats (date(start_date, "unixepoch"))""")
self.conn.commit()
LOG.info("Indexes created")
def import_bikepoints(self):
LOG.info("Importing bikepoints")
points = json.loads(requests.get("https://api.tfl.gov.uk/BikePoint").text)
points = list(map(lambda p: (p['id'], p['commonName'], p['lat'], p['lon']), points))
self.conn.executemany("INSERT OR IGNORE INTO bike_points VALUES (?, ?, ?, ?)", points)
self.conn.commit()
LOG.info("Bikepoints imported")
def is_file_already_imported(self, etag):
rows = self.conn.execute("SELECT * FROM read_files WHERE etag LIKE ?", (etag,)).fetchall()
@ -81,6 +110,8 @@ class BikeDatabase:
LOG.warning(f"Skipping import of {export_file.path}")
return
cursor = self.conn.cursor()
os.makedirs(os.path.dirname(export_file.path), exist_ok=True)
LOG.info(f"DOWNLOADING... {export_file.download_url} to {export_file.path}")
urllib.request.urlretrieve(export_file.download_url, export_file.path)
@ -119,8 +150,8 @@ class BikeDatabase:
LOG.error(f"Key Error {e} on line {entry}")
return
LOG.info(f"Writing {len(mapped)} entries to DB")
self.cursor.executemany("INSERT INTO usage_stats VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)", mapped)
self.cursor.execute("INSERT INTO read_files VALUES (?, ?)", (export_file.path, export_file.etag))
cursor.executemany("INSERT INTO usage_stats VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)", mapped)
cursor.execute("INSERT INTO read_files VALUES (?, ?)", (export_file.path, export_file.etag))
self.conn.commit()
LOG.info(f"Finished import of {export_file.path}")
os.remove(export_file.path)
@ -130,16 +161,17 @@ class BikeDatabase:
def main():
all_files = get_online_files_list(subdir_filter="usage-stats", file_extension_filter=".csv")
db = BikeDatabase()
db.init_database()
# General DB init
BikeDatabase().init_database()
# Download and import opendata from S3 bucket
os.makedirs(TMP_DIR, exist_ok=True)
os.chdir(TMP_DIR)
LOG.info("Switching into tmp dir")
import_tasks = []
with ThreadPoolExecutor(3) as executor:
with ThreadPoolExecutor(1) as executor:
for file in all_files:
db = BikeDatabase()
import_tasks.append(
@ -152,6 +184,11 @@ def main():
shutil.rmtree(TMP_DIR)
LOG.info("Deleted temp dir")
# Import Bikepoints
BikeDatabase().import_bikepoints()
# Create search-index for faster querying
BikeDatabase().create_indexes()
if __name__ == "__main__":
main()