Published: Oct 10, 2019 by K. E. Claytor
Here’s the recording:
I think it’s a decent match for research systems. You can get up to speed fairly quickly without a background in database systems, and the time-series first aspect of it is good for a wide range of datasets.
- Easy to get started with many clients across most languages and a HTTP API.
- Telegraf also records system metrics
- Chronograf integration gives you visualization with low overhead.
- NoSQL means you can change your schema as the experiment evolves (and you don’t even need a schema to get started).
- Memory issues on IoT platforms (I’m using a snickerdoodle board).
- Runs out of memory on start (supposedly fixed v1.7.8?)
- Runs out of memory on compactions (fixed by moving to smaller shards)
- Takes a long time to boot up and read lots of shards.
- Python client doesn’t read into Pandas tables very well (it can write from them well).
I usually export from chronograf into a .csv file and then
- Will drop points if the type doesn’t match (explicitly cast python variables).