UI Error: Connection Closed After Periods of Inactivity | Arize AI Community

Arize AI Community Icon

UI Error: Connection Closed After Periods of Inactivity | Arize AI Community

RunLLM
APP
·
It seems likely that the error you're encountering is related to a connection timeout or a closed connection with the PostgreSQL database. This can happen if the connection is idle for too long and the database server closes it. One way to address this issue is by configuring a connection pool with appropriate timeout settings to keep the connection alive or to automatically reconnect when needed. My understanding is that you might want to adjust the connection pool settings in your SQLAlchemy configuration. You can set parameters like `pool_recycle` to a lower value to ensure that connections are recycled before they are closed by the server. Additionally, you can use `pool_pre_ping` to check the connection's health before using it, which can help in automatically reconnecting if the connection is closed. Would you like more details on how to configure these settings? Sources: - <https://github.com/Arize-ai/phoenix/issues/7193|https://github.com/Arize-ai/phoenix/issues/7193> - <https://github.com/Arize-ai/phoenix/issues/6296|https://github.com/Arize-ai/phoenix/issues/6296>
💡 Hint: Mention RunLLM in the thread for followups.
Xander S.
·
Hey Zhuyi X., what page are you encountering this error on?
Zhuyi X.
·
home page
Xander S.
·
Thanks Zhuyi X.. That's pretty odd. Do you hit the error when you first try to interact with Phoenix after a period of idle time?
Xander S.
·
Or does the error appear by itself with no interaction?
Zhuyi X.
·
Do you hit the error when you first try to interact with Phoenix after a period of idle time?
this is the case
👍1
Xander S.
·
Thanks for reporting Zhuyi X.. I'm not sure I fully understand why this issue is happening. It seems like when you hit Postgres for the first time after a period of idleness, the connection tends to drop. I've filed an issue here. It looks like this particular query does not actually need to hit the database, so I've simplified the query in this PR. Not sure if this will resolve the issue, but I will publish a new version of Phoenix and let you try it out.
1
Xander S.
·
The change will be released shortly as part of v8.28.1. Please let us know if that resolves the issue!
Zhuyi X.
·
Thanks. Btw, more context, I host my phoenix inside k8s, I wonder if it could be due to my tab idling for long and a pod restart happens. I wonder if pod restart will lead to some unwanted effect related to the connections/session.
Xander S.
·
I'm not clear how that would be the cause of this particular issue. The fact that the UI receives an error message indicates that the browser is able to communicate with an individual Phoenix pod. It appears that it is the connection between the Phoenix pod and the Postgres instance that is problematic in this case.
💡1
Xander S.
·
Maybe I am missing something.