schedview Reports#

Introduction#

schedview can be used to generate a any of several reports. Most of these reports are built from jupyter notebooks, either using Times Square or nbconvert.

Times Square schedview notebooks#

The Times Square schedview notebooks can be found at https://usdf-rsp.slac.stanford.edu/times-square/github/lsst/schedview_notebooks/nightly/scheduler-nightsum. Times Square reports are generated on demand, but cached after the first time they are executed. Because the schedview notebooks can be slow to execute, this is sometimes incovenient, or can fail altogether.

Pre-generated reports#

Introduction#

To avoid requiring humans to wait for notebooks to be executed, a cron job at the USDF uses submits a set of batch jobs that use nbconvert to create a set of html pages. These scripts are found in the batch directory of the schedview_notebooks repository: batch/prenight.sh and batch/scheduler_nightsum.sh. Each of these scripts builds an environment in which to convert the notebook, executes the conversion, and updates a corresponding index it include the new reports. There are currently two indexes of reports. One index lists publicly accessible, which are served from https://s3df.slac.stanford.edu/data/rubin/sim-data/schedview/reports/. Currently, the public reports are limited to a brief nigth summary.

Additional reports, currently requiring logging to the the USDF, can be found at https://usdf-rsp-int.slac.stanford.edu/schedview-static-pages/.

Job sumbission#

The cron job runs the reports with the following entries:

15 5 * * * /opt/slurm/slurm-curr/bin/sbatch /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks/batch/scheduler_nightsum.sh 2>&1 >> /sdf/data/rubin/shared/scheduler/schedview/scheduler_nightsum/scheduler_nightsum.out
30 7 * * * /opt/slurm/slurm-curr/bin/sbatch /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks/batch/prenight.sh 2>&1 >> /sdf/data/rubin/shared/scheduler/schedview/prenight/prenight.out

The environment#

The conda environment provided in the jupyter aspect of the USDF RSP, and therefore alse Time Square, differs somewhat from the “lsst” environment available on the development nodes.

To get these batch scripts to execute as they would on a jupyter aspect notebook or on Times Square, the conda environment there needs to be duplicated somewhere it can be activated from the slurm job.

Begin by starting a USDF notebook-aspect jupyter lab, making careful note of which image you select.

From within that jupyterlab, start a terminal, and record the contents of the environment provided. In the example used below, this was w2025_36, and the schedview_notebooks repository was checked-out into /sdf/data/rubin/user/neilsen/devel/schedview_notebooks .

So, recording the conda environment we want to replicate:

setup lsst_sitcom
cd /sdf/data/rubin/user/neilsen/devel/schedview_notebooks/batch
conda list --explicit > like_rsp_w2025_36_spec.txt

Then log into a USDF development node, go to the directory where you saved the spec file above, and create an environment in the shared scheduler space:

cd /sdf/data/rubin/user/neilsen/devel/schedview_notebooks/batch
source /sdf/group/rubin/sw/w_latest/loadLSST.sh
conda create --prefix /sdf/data/rubin/shared/scheduler/envs/like_rsp_w2025_36 --file  like_rsp_w2025_36_spec.txt

The batch script that generates the prenight briefing reeport requeires a version of schedview newer than currently on the RSP, so make a variant with the new schedview added:

conda create \
    --prefix /sdf/data/rubin/shared/scheduler/envs/prenight_like_rsp_w2025_36 \
    --clone /sdf/data/rubin/shared/scheduler/envs/like_rsp_w2025_36
conda activate /sdf/data/rubin/shared/scheduler/envs/prenight_like_rsp_w2025_36
pip install git+https://github.com/lsst/schedview.git@v0.19.0.dev1

Finally, update the bash scripts that need it (batch/prenight.sh and batch/scheduler_nightsum.sh), for example:

source /sdf/group/rubin/sw/w_latest/loadLSST.sh
conda activate /sdf/data/rubin/shared/scheduler/envs/like_rsp_w2025_36

Even though we aren’t using the environment provided by the source of loadLSST.sh above, it’s still needed to get conda into our path.

Updating version of schedview_notebooks used by the cron job#

The scripts submitted by the cron job supplied above use the version of schedview in /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks, which is itsef a link to a directory for a specific version, e.g. /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks-v0.1.0.

To tag and install a new version to be used, start by deciding on a tag. Get sorted existing tags with:

curl -s https://api.github.com/repos/lsst/schedview_notebooks/tags \
    | jq -r '.[].name' \
    | egrep '^v[0-9]+.[0-9]+.[0-9]+.*$' \
    | sort -V

Make and push a new tag (with the base of a checked-out schedview_notebooks as the current working directory):

git tag v0.1.0.dev2
git push origin tag v0.1.0.dev2

Then install it in /sdf/data/rubin/shared/scheduler/packages:

git archive \
    --format=tar \
    --prefix schedview_notebooks-v0.1.0.dev2/ v0.1.0.dev2 \
    | tar -x --directory /sdf/data/rubin/shared/scheduler/packages

Replace the symlink to point to your new one:

if test -L /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks ; then rm /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks ; fi
ln -s /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks-v0.1.0.dev2 /sdf/data/rubin/shared/scheduler/packages/schedview_notebooks

Hand generation of reports#

The schedview_notebooks github repository contains the notebooks themselves. Execution of the notebooks requires installation schedview and its dependencies, following the schedview intallation documentation. These can be run directly with a user’s jupyter instance, or converted to html using nbconvert. The notebooks themselves containt documentation about how to do this conversion. The general pattern followed by these instructions is:

  1. Setup the python environment.

  2. Assign a set of environment variables. These variables map to the parameters used when the notebooks are run using Times Square.

  3. Call nbconvert with a command that looks similar to this:

    jupyter nbconvert \
--to html \
--execute \
--no-input \
--ExecutePreprocessor.kernel_name=python3 \
--ExecutePreprocessor.startup_timeout=3600 \
--ExecutePreprocessor.timeout=3600 \
whatever_notebook.ipynb