error ❌ tRPC failed on public.checkUpdate: Failed to fetch or json parse the latest releases Failed to fetch or json parse the latest releases

[FaresKi]

Hi!
Currently self hosting Langfuse on K8S, we noticed the following error message on the running server:

2024-09-30T08:16:01.768Z error ❌ tRPC failed on public.checkUpdate: Failed to fetch or json parse the latest releases Failed to fetch or json parse the latest releases
TRPCError: Failed to fetch or json parse the latest releases
at /app/web/.next/server/pages/api/trpc/[trpc].js:379:4188
at process.processTicksAndRejections (node:internal/process/task_queues:95:5)
at async resolveMiddleware (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/index.mjs:421:30)
at async callRecursive (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/index.mjs:451:32)
at async file:///app/node_modules/.pnpm/@baselime+trpc-opentelemetry-middleware@0.1.2_@opentelemetry+api@1.8.0_@trpc+server@10.45.2/node_modules/@baselime/trpc-opentelemetry-middleware/dist/index.js:43:22
at async callRecursive (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/index.mjs:451:32)
at async resolve (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/index.mjs:481:24)
at async inputToProcedureCall (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/resolveHTTPResponse-2fc435bb.mjs:46:22)
at async Promise.all (index 0)
at async resolveHTTPResponse (file:///app/node_modules/.pnpm/@trpc+server@10.45.2/node_modules/@trpc/server/dist/resolveHTTPResponse-2fc435bb.mjs:182:37)

When using langfuse as a callback handler for my LLM operations, I get this on my LLM app:

ERROR - Unexpected error occurred. Please check your request and contact support: https://langfuse.com/support.

Is this a temporary bug? Like redeploying the app would solve the issue?

[dosubot[bot]]

Yes, the error "tRPC failed on public.checkUpdate: Failed to fetch or json parse the latest releases" could potentially be a temporary issue that might be resolved by redeploying the Langfuse app on Kubernetes (K8S). This error occurs when the application fails to fetch or parse the JSON response from the https://langfuse.com/api/latest-releases endpoint. This could be due to a temporary network issue, a problem with the external API, or a transient error in the application environment. Redeploying the app might resolve such transient issues. Additionally, ensure that the GitHub API requests are properly authenticated and that the cache mechanism is functioning correctly ^[1][2].

To continue talking to Dosu, mention @dosu.

[FaresKi]

thanks marc, it's much appreciated

[FaresKi]

@dosu hi
coming back with more information.
the server seems to log this error:

Invalid tokenizer config for model cm1ush7dh0001kw2zmt0kkq0e: {"tokensPerName":1,"tokenizerModel":"gpt-4o-2024","tokensPerMessage":3}, {"issues":[{"code":"custom","message":"Unknown tiktoken model","path":["tokenizerModel"]}],"name":"ZodError"}

for some context we don't use the standard models, but for pricing reasons we use the openai tokenizer.
we created a custom model in langfuse and locally it's working fine, but not in the remote server...
Hopefully that provides more answers in regards to where the error could be

[dosubot[bot]]

The error "Invalid tokenizer config for model cm1ush7dh0001kw2zmt0kkq0e" occurs because the tokenizerModel specified, "gpt-4o-2024", is not recognized by the tiktoken package used for tokenization. This leads to a ZodError indicating an unknown tokenizerModel ^[1].

This issue likely arises because the remote server does not have the custom model definition properly configured, or the tiktoken package on the server does not support the specified model. Locally, it might work if the local environment has the correct configuration or a different version of the tiktoken package that recognizes the model.

To resolve this, ensure that the custom model definition is correctly set up on the remote server, and verify that the tiktoken package version used on the server supports the specified tokenizerModel. If necessary, update the model definition to use a recognized tokenizerModel or adjust the server's tiktoken package to match the local environment ^[1].

To continue talking to Dosu, mention @dosu.

[FaresKi]

Ahhh it's gpt-4o not gpt-4o-2024!!!

[marcklingen]

great to see that dosu helped resolve this!