Files
cram-luftdaten/toLineProtocol.py
2019-09-07 00:54:43 +02:00

85 lines
3.0 KiB
Python
Executable File

#!/usr/bin/env python3
#
# Converts CSV exports from https://www.madavi.de/sensor/csvfiles.php to
# InfluxDB LineProtocol https://docs.influxdata.com/influxdb/v1.2/write_protocols/line_protocol_reference/
#
# Note: timestamps are in seconds, therefore precision "s" needs to be set
# when writing, see https://docs.influxdata.com/influxdb/v1.2/tools/api/#write
#
# Settings:
#
# Please set
# * SENSOR_ID
# * INFLUXDB_DATABASE
# in your environment
# CSV file specs
#
# | DB Field | DB field | CSV Column | CSV Format
# |--------------------|----------|--------------------|--------------------------
# | time | time | Time | 2017/05/19 00:00:11 (UTC)
# | ? | ? | durP1 |
# | ? | ? | ratioP1 |
# | ? | ? | P1 |
# | ? | ? | durP2 |
# | ? | ? | ratioP2 |
# | ? | ? | P2 |
# | SDS_P1 | field | SDS_P1 | 10.36
# | SDS_P2 | field | SDS_P2 | 9.50
# | PMS_P1 | field | PMS_P1 |
# | PMS_P2 | field | PMS_P2 |
# | temperature | field | Temp | 16.00
# | humidity | field | Humidity | 79.00
# | BMP_temperature | field | BMP_temperature |
# | BMP_pressure | field | BMP_pressure |
# | BME280_temperature | field | BME280_temperature |
# | BME280_humidity | field | BME280_humidity |
# | BME280_pressure | field | BME280_pressure |
# | samples | field | Samples | 828799
# | min_micro | field | Min_cycle | 172
# | max_micro | field | Max_cycle | 25198
# | signal | field | Signal | -91
# | node | tag | N/A filename? | e.g. esp8266-16229960
import os
import sys
import csv
from datetime import datetime, timedelta
def get_timestamp(timestr):
naive_dt = datetime.strptime(timestr, '%Y/%m/%d %H:%M:%S');
utc = (naive_dt - datetime(1970, 1, 1)) / timedelta(seconds=1)
return int(utc)
SENSOR_ID = os.environ.get('SENSOR_ID', '16229960')
DATABASE = os.environ.get('INFLUXDB_DATABASE', 'sensors')
NODE = 'esp8266-' + SENSOR_ID
NAME_MAP = {
'Humidity': 'humidity',
'Max_cycle': 'max_micro',
'Samples': 'samples',
'Min_cycle': 'min_micro',
'Signal': 'signal',
'Temp': 'temperature'
}
READER = csv.DictReader(sys.stdin, delimiter=';')
for row in READER:
measurements = []
for header, value in row.items():
if header == 'Time' or not value:
continue
measurements.append('{0}={1}'.format(NAME_MAP.get(header, header), value))
values = {
'database': DATABASE,
'node': NODE,
'measurements': ','.join(measurements),
'time': get_timestamp(row['Time'])
}
print('{database},node={node} {measurements} {time}'.format(**values))