Arize AI Community Icon

Home
Events

Arize AX Releases
Arize AX Support
Arize News
Arize Observe
Discussions
Introductions
Phoenix Support

Powered by Tightknit

Phoenix Support

Understanding Persistent Storage in Phoenix for Chatbot Deployment | Arize AI Community

Phoenix Support

Understanding Persistent Storage in Phoenix for Chatbot Deployment

·Apr 12, 2024 11:25 AM

Hi, I'm looking into deploying phoenix as part of deployment stack for our chatbot and I'm confused about persistent storage. Is that correct that there is no way currently to store data between phoenix restarts?

5 comments

· Sorted by Oldest

Mikyo
·
Arturs V. this is our current milestone but as you pointed out container instances don’t natively preserve data right now. You can still download the data https://docs.arize.com/phoenix/tracing/how-to-tracing/save-and-load-traces but this is the best we have at the current juncture. Thanks again for using phoenix and persistence is coming very very soon.
Arturs V.
·
Nice, that means I can create a wrapper around it, dockerise it and periodically save the changes and mount the volume to be saved in and next time just check if something is inside the directory and import it.
Mikyo
·
Not quite that elegant yet but yeah if you save it off we will give you facilities to send them back. cc Dustin N. who's working on the send_traces utility.
👍2
Arturs V.
·
I ended up writing custom code that periodically saves traces to disk and when starts the phoenix then it takes all the trace files(panda data frames) combines and imports them. Using the rest API, did not work since it was failing to serialize data frames, because of different data types for same column. To solve that I used pickle instead of pyarrow.
Mikyo
·
Wow, great work!