You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Add HPC environment module and SPARKCTL_SETTINGS_FILE support
Add a ready-to-deploy environment module (TCL and Lua flavors) under
hpc/environment_module/ so HPC users can run:
module load sparkctl
sparkctl configure --start
The module activates a shared pre-installed virtualenv, sets
SPARKCTL_SETTINGS_FILE, and puts spark-submit/pyspark on PATH. It reads
the Spark/Java/Hadoop binary paths from the shared sparkctl.toml so that
settings file is the single source of truth (TCL parses it natively; Lua
parses it via awk since Lmod's sandbox lacks reliable file I/O).
Wire up the SPARKCTL_SETTINGS_FILE environment variable in config.py
(it was already documented in the deploy guide but never implemented).
Settings files now load in increasing precedence: ~/.sparkctl.toml, then
$SPARKCTL_SETTINGS_FILE, then ./.sparkctl.toml.
Document the module and the env var in the deploy guide, and bump the
version to 0.4.1 (first release supporting SPARKCTL_SETTINGS_FILE).
Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>