Compare commits

..

No commits in common. "eae785064efb9bbbd3556400a5b2262c8c5c30ba" and "2adef5aa100774c1ea67242a205130ede52a51aa" have entirely different histories.

2 changed files with 97 additions and 163 deletions

View File

@ -6,168 +6,105 @@ import io
import zipfile import zipfile
import datetime import datetime
from config.db import db as config_db from config.db import db as config_db
import threading
class DwdScraper:
def __init__(self, worker=1):
self._worker = worker
self._tasks = {}
self._request_new_task = threading.Event()
self._close = True
def run(self):
for n in range(self._worker):
threading.Thread(target=self._run_scraper, kwargs={"id": n}, daemon=True).start()
while self._close:
self._request_new_task.wait()
self._request_new_task.clear()
print("Checking for waiting workers")
for id, t in list(self._tasks.items()):
if t["dwd_id"] is None:
print("Worker {} needs a job".format(id))
dwd_id = self.get_station_to_update()
if dwd_id is not None:
print("Worker {} got {}".format(id, dwd_id))
self._tasks[id]["dwd_id"] = dwd_id
self._tasks[id]["wait_for_dwd_id"].set()
def _run_scraper(self, id):
print("Worker {} spawned".format(id))
while self._close:
self._tasks[id] = {
"wait_for_dwd_id": threading.Event(),
"dwd_id": None
}
self._request_new_task.set()
print(id, "Waits for new task")
self._tasks[id]["wait_for_dwd_id"].wait()
print(id, "Got task {}".format(self._tasks[id]["dwd_id"]))
self._do_scrape(self._tasks[id]["dwd_id"])
def _do_scrape(self, dwd_id):
last_station = dwd_id
r = requests.get('https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily/kl/recent/tageswerte_KL_' + str(last_station) + '_akt.zip', stream=True)
print(dwd_id, "downloaded file")
r.raise_for_status()
zip = zipfile.ZipFile(io.BytesIO(r.content))
files = zip.namelist()
files_climate = [f for f in files if f.startswith("produkt_klima_")]
with psycopg2.connect(config_db["uri"]) as conn:
with conn.cursor() as cur:
cur.execute("SELECT id, name, lat, lon, dwd_id, dwd_last_update FROM stations WHERE dwd_id = %s;", [last_station])
current_station = cur.fetchone()
curr_station_id = current_station[0]
curr_station_dwd_id = current_station[4]
is_first_line = True
count = 0
for line in zip.open(files_climate[0]):
l = line.decode('utf-8').strip()
if is_first_line:
is_first_line = False
continue
count += 1
b = {}
b["dwd_id"], l = l.strip().split(";", 1)
b["dwd_id"] = str(b["dwd_id"])
b["date"], l = l.strip().split(";", 1)
b["date"] = str(b["date"])
b["date"] = datetime.date(int(b["date"][0:4]), int(b["date"][4:6]), int(b["date"][6:8]))
b["qn_3"], l = l.strip().split(";", 1)
b["qn_3"] = self.cleanup_value(int(b["qn_3"]))
b["fx"], l = l.strip().split(";", 1)
b["fx"] = self.cleanup_value(float(b["fx"]))
b["fm"], l = l.strip().split(";", 1)
b["fm"] = self.cleanup_value(float(b["fm"]))
b["qn_4"], l = l.strip().split(";", 1)
b["qn_4"] = self.cleanup_value(int(b["qn_4"]))
b["rsk"], l = l.strip().split(";", 1)
b["rsk"] = self.cleanup_value(float(b["rsk"]))
b["rskf"], l = l.strip().split(";", 1)
b["rskf"] = self.cleanup_value(int(b["rskf"]))
b["sdk"], l = l.strip().split(";", 1)
b["sdk"] = self.cleanup_value(float(b["sdk"]))
b["shk_tag"], l = l.strip().split(";", 1)
b["shk_tag"] = self.cleanup_value(float(b["shk_tag"]))
b["nm"], l = l.strip().split(";", 1)
b["nm"] = self.cleanup_value(float(b["nm"]))
b["vpm"], l = l.strip().split(";", 1)
b["vpm"] = self.cleanup_value(float(b["vpm"]))
b["pm"], l = l.strip().split(";", 1)
b["pm"] = self.cleanup_value(float(b["pm"]))
b["tmk"], l = l.strip().split(";", 1)
b["tmk"] = self.cleanup_value(float(b["tmk"]))
b["upm"], l = l.strip().split(";", 1)
b["upm"] = self.cleanup_value(float(b["upm"]))
b["txk"], l = l.strip().split(";", 1)
b["txk"] = self.cleanup_value(float(b["txk"]))
b["tnk"], l = l.strip().split(";", 1)
b["tnk"] = self.cleanup_value(float(b["tnk"]))
b["tgk"], l = l.strip().split(";", 1)
b["tgk"] = self.cleanup_value(float(b["tgk"]))
cur.execute("SELECT id FROM climate WHERE station = %s AND date = %s; ", [curr_station_id, b["date"].isoformat()])
if cur.rowcount == 0:
cur.execute("INSERT INTO climate (station, date, qn_3, fx, fm, qn_4, rsk, rskf, sdk, shk_tag, nm, vpm, pm, tmk, upm, txk, tnk, tgk) VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s);",
[curr_station_id, b["date"].isoformat(), b["qn_3"], b["fx"], b["fm"], b["qn_4"], b["rsk"], b["rskf"], b["sdk"], b["shk_tag"], b["nm"], b["vpm"], b["pm"], b["tmk"], b["upm"], b["txk"], b["tnk"], b["tgk"]])
if count % 20 == 0:
print(dwd_id, "still updating")
print(dwd_id, "finished import")
cur.execute("UPDATE stations SET dwd_last_update = %s WHERE dwd_id = %s;", [datetime.datetime.today().isoformat(), last_station])
conn.commit()
print(dwd_id, "updated last change data")
def get_station_to_update(self):
"""
Returns DWD ID of station that should get updated soon.
"""
with psycopg2.connect(config_db["uri"]) as conn:
with conn.cursor() as cur:
cur.execute("SELECT dwd_id FROM stations ORDER BY dwd_last_update ASC LIMIT %s", [self._worker + 1])
stations = cur.fetchall()
#print(stations)
for station in stations:
if station[0] not in [t["dwd_id"] for id, t in self._tasks.items()]:
return station[0]
def cleanup_value(v):
if int(v) == -999:
return None return None
def cleanup_value(self, v): return v
if int(v) == -999:
return None
return v conn = psycopg2.connect(config_db["uri"])
cur = conn.cursor()
if __name__ == "__main__": cur.execute("SELECT min(dwd_last_update) FROM stations;")
DwdScraper(worker=8).run() last_date = cur.fetchall()[0][0]
print(last_date)
if not last_date is None:
cur.execute("SELECT id, name, lat, lon, dwd_id, dwd_last_update FROM stations WHERE dwd_last_update = %s LIMIT 1;", [last_date])
else:
cur.execute("SELECT id, name, lat, lon, dwd_id, dwd_last_update FROM stations WHERE dwd_last_update IS NULL LIMIT 1;")
last_station = cur.fetchone()
print(last_station)
curr_station_id = last_station[0]
curr_station_dwd_id = last_station[4]
print(curr_station_dwd_id)
r = requests.get('https://opendata.dwd.de/climate_environment/CDC/observations_germany/climate/daily/kl/recent/tageswerte_KL_' + str(curr_station_dwd_id) + '_akt.zip', stream=True)
print(r.url)
if r.status_code == 200:
zip = zipfile.ZipFile(io.BytesIO(r.content))
files = zip.namelist()
print(files)
files_climate = [f for f in files if f.startswith("produkt_klima_")]
print(files_climate)
buffer = []
is_first_line = True
for line in zip.open(files_climate[0]):
l = line.decode('utf-8').strip()
if is_first_line:
is_first_line = False
continue
b = {}
b["dwd_id"], l = l.strip().split(";", 1)
b["dwd_id"] = str(b["dwd_id"])
b["date"], l = l.strip().split(";", 1)
b["date"] = str(b["date"])
b["date"] = datetime.date(int(b["date"][0:4]), int(b["date"][4:6]), int(b["date"][6:8]))
b["qn_3"], l = l.strip().split(";", 1)
b["qn_3"] = cleanup_value(int(b["qn_3"]))
b["fx"], l = l.strip().split(";", 1)
b["fx"] = cleanup_value(float(b["fx"]))
b["fm"], l = l.strip().split(";", 1)
b["fm"] = cleanup_value(float(b["fm"]))
b["qn_4"], l = l.strip().split(";", 1)
b["qn_4"] = cleanup_value(int(b["qn_4"]))
b["rsk"], l = l.strip().split(";", 1)
b["rsk"] = cleanup_value(float(b["rsk"]))
b["rskf"], l = l.strip().split(";", 1)
b["rskf"] = cleanup_value(int(b["rskf"]))
b["sdk"], l = l.strip().split(";", 1)
b["sdk"] = cleanup_value(float(b["sdk"]))
b["shk_tag"], l = l.strip().split(";", 1)
b["shk_tag"] = cleanup_value(float(b["shk_tag"]))
b["nm"], l = l.strip().split(";", 1)
b["nm"] = cleanup_value(float(b["nm"]))
b["vpm"], l = l.strip().split(";", 1)
b["vpm"] = cleanup_value(float(b["vpm"]))
b["pm"], l = l.strip().split(";", 1)
b["pm"] = cleanup_value(float(b["pm"]))
b["tmk"], l = l.strip().split(";", 1)
b["tmk"] = cleanup_value(float(b["tmk"]))
b["upm"], l = l.strip().split(";", 1)
b["upm"] = cleanup_value(float(b["upm"]))
b["txk"], l = l.strip().split(";", 1)
b["txk"] = cleanup_value(float(b["txk"]))
b["tnk"], l = l.strip().split(";", 1)
b["tnk"] = cleanup_value(float(b["tnk"]))
b["tgk"], l = l.strip().split(";", 1)
b["tgk"] = cleanup_value(float(b["tgk"]))
#print(b)
print(curr_station_id, b["date"].isoformat())
cur.execute("SELECT id FROM climate WHERE station = %s AND date = %s; ", [curr_station_id, b["date"].isoformat()])
if cur.rowcount == 0:
cur.execute("INSERT INTO climate (station, date, qn_3, fx, fm, qn_4, rsk, rskf, sdk, shk_tag, nm, vpm, pm, tmk, upm, txk, tnk, tgk) VALUES (%s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s, %s);",
[curr_station_id, b["date"].isoformat(), b["qn_3"], b["fx"], b["fm"], b["qn_4"], b["rsk"], b["rskf"], b["sdk"], b["shk_tag"], b["nm"], b["vpm"], b["pm"], b["tmk"], b["upm"], b["txk"], b["tnk"], b["tgk"]])
print("imported")
else:
print("ignored")
cur.execute("UPDATE stations SET dwd_last_update = %s WHERE id = %s;", [datetime.datetime.today().isoformat(), curr_station_id])
conn.commit()
cur.close()
conn.close()

View File

@ -76,13 +76,10 @@ if r.status_code == 200:
cur = conn.cursor() cur = conn.cursor()
for b in buffer: for b in buffer:
print("importing", b["station_id"])
cur.execute("SELECT * FROM stations WHERE dwd_id LIKE %s;", [str(b["station_id"])]) cur.execute("SELECT * FROM stations WHERE dwd_id LIKE %s;", [str(b["station_id"])])
if cur.rowcount == 0: if cur.rowcount == 0:
print("\t", "new station")
cur.execute("INSERT INTO stations (name, lat, lon, dwd_id, state, sea_level) VALUES (%s, %s, %s, %s, %s, %s);", [b["name"], b["lat"], b["lon"], b["station_id"], b["state"], b["sea_level"]]) cur.execute("INSERT INTO stations (name, lat, lon, dwd_id, state, sea_level) VALUES (%s, %s, %s, %s, %s, %s);", [b["name"], b["lat"], b["lon"], b["station_id"], b["state"], b["sea_level"]])
else: else:
print("\t", "update station")
cur.execute("UPDATE stations SET name = %s, lat = %s, lon = %s, state = %s, sea_level = %s WHERE dwd_id LIKE %s", [b["name"], b["lat"], b["lon"], b["state"], b["sea_level"], b["station_id"]]) cur.execute("UPDATE stations SET name = %s, lat = %s, lon = %s, state = %s, sea_level = %s WHERE dwd_id LIKE %s", [b["name"], b["lat"], b["lon"], b["state"], b["sea_level"], b["station_id"]])
conn.commit() conn.commit()