Thank you for you question and apologise for slight delay.
We needed to take some time to complete our internal investigation and believe we’ve identified the root cause of the tmp file buildup.
When the GCS upload fails with “Unexpected end of file from server” (a transient network error), the connector retries the operation as expected. However, we believe there is a bug where the local staging file from the failed attempt is not properly cleaned up — each retry creates a new staging file while the previous one remains on disk. With a high number of tasks and continuous upload failures, these orphaned files accumulate and eventually exhaust ephemeral storage.
Our team worked to patch this behaviour and therefore we released Stream Reactor 11.6.1, which should help resolve this issue.