The issue is now resolved.
/scratch remains under heavy load from thousands of computing jobs, so performance may vary. We’ll keep monitoring things.
Please make sure to take advantage of the node-local
$L_SCRATCH storage location whenever possible, especially when sharing data across multiple nodes is not necessary (more information is available at https://www.sherlock.stanford.edu/docs/storage/)
We’re currently working on an issue with
Symptoms are hanging commands when accessing certain parts of the file system, jobs stuck in COMPLETING/CG state, and potential errors when logging in.
We expect the issue to be resolved shortly.