mirror of
https://github.com/element-hq/synapse.git
synced 2026-09-25 22:04:00 +00:00
Erik Johnston
818f2b25bb
We sometimes see a lot of `ERROR` logs like the following: ``` Closing scope Scope<... master.write_bytes_to_request> which is not the currently-active one None ``` This error is generated by `opentracing.Scope` checking that it is the "active" one. Synapse tracks "active" spans via logcontexts, so this indirectly asserts that the scope is closed in the context it was opened in. However, the producer methods are often called from the reactor and therefore withing the sentinel logcontext, which produces the error above. This specifically happens when the producer tries to write large responses but gets paused, and then later resumes. The fix is to simply use `Span` directly, rather than scopes. `Span` does not perform the checks. I noticed this when deploying #19979, though it is unrelated.