Download tasks/format-code-task-000055/instruction.md from FineEnvs/MiMo-V2.6-RL-harbor-code: direct link, hf CLI and curl.
- Browser
- Download file 1.47 kB
-
https://huggingface.co/datasets/FineEnvs/MiMo-V2.6-RL-harbor-code/resolve/main/tasks/format-code-task-000055/instruction.md
- Command line
-
hf download hf://datasets/FineEnvs/MiMo-V2.6-RL-harbor-code/tasks/format-code-task-000055/instruction.md
-
curl -L -o instruction.md https://huggingface.co/datasets/FineEnvs/MiMo-V2.6-RL-harbor-code/resolve/main/tasks/format-code-task-000055/instruction.md
Fix the following issue:
RUM events being ingested without a sessionId
While looking at the RUM events our app sends to the Datadog intake, I noticed a non-trivial number of them are missing a sessionId (the field is just absent / undefined on the event). These same events still have a view id and look otherwise valid, they just have no session attached, which makes them useless for session-level analysis.
After some digging I managed to reproduce it with the following setup:
- Open our app in two tabs, both initialized with the browser RUM SDK. The session is initially not sampled / not tracked.
- Wait for the session to expire.
- In tab A, do something that causes the SDK to renew the session, and this time the new session lands in the tracked bucket.
- Switch to tab B (do not interact with it — don't click, don't navigate, just let background things like XHRs, long tasks, errors etc. happen).
Tab B keeps sending RUM events to the intake, and those are the ones showing up without a sessionId. Tab B never had a chance to react to the renewed session — from its point of view it's still on the last view of the now-expired session — yet events from it keep flowing through.
I'd expect the SDK to simply not send RUM events when it can't attach them to a real session. A tab that hasn't seen any actual user interaction since the session was renewed shouldn't be silently producing orphan events in the backend.