Compute Jobs – Excess Memory Usage
When running a Job in the Small Job Domain, if a user allocates more than the allowed amount of RAM (approx 40GB) the Docker Container will crash and hang until the Job Timeout is met. the user will need to manually cancel the Job.
Compute Jobs – Timeout shows as Cancelled
Both the Small Jobs and Large Jobs Domains have processing timeouts. When a Job times out, the Jobs UI will show this job as “cancelled”, but the description will still show it as timed-out.
Rendering on Smartphones
The SciServer UI does not render very well on smartphones, some of the dialog entry forms are large and don’t fit well. Larger tablets are better, but the recommendation is to run SciServer from a computer.
Installing Tensorflow in a Compute Container
Installing tensorflow with ‘pip install tensorflow’ or ‘conda install tensorflow’ will not work due to https://github.com/tensorflow/tensorflow/issues/17411 . It used to work in the 1.10.x version of Compute, with older version of Tensorflow. Tensorflow 1.5 does work in our system, and we are working on a machine learning images with a special build of the newest version of Tensorflow for our system.