Hello, I write time series data to influxDB using UDP protocol, and the configuration bind the database to the specified port. However can I set the database name in UDP packet so that UDP listener will write the data to the corresponding database? I would like to create database in dynamic and part the data in year, say data_2017, data_2018 etc.
@jacktang Why would you want to separate the databases by years? This would make the data more difficult to query and separates it needlessly. Also while the UDP input is slightly more performant the the HTTP API, that performance difference is pretty trivial. Unless you are writing upwards of ~500,000 points / second there is no need to use UDP.
I am going to some high frequent finance tick data into influxdb and the data safety is the first thing I should consider. I have tested one-year data volume and it works fine. I don’t find some benchmark like how much data can be stored in influxdb and the related performance. If you know that, could you please tell me?
More, I set up database hashed by year, it makes archiving and swapping out historical datum easy(?). I choose UDP protocol, because of 1) performance, 2) setting up master-slave quickly.
@jacktang We shard data by time already so splitting it into separate databases has no performance impact and isn’t any easier for swapping out historical data. I think storing the data in a single database would be easier from an administrative point of view.
The quantity of data stored has little performance impact, especially on realtime queries. We have folks with 10TB+ installations.
If data safety is a priority, I would suggest using
udp. The performance gain doesn’t justify the the likelihood of dropped packets.