Details
-
Type:
RFC
-
Status: Implemented
-
Resolution: Done
-
Component/s: DM
-
Labels:None
Description
Currently, ap_pipe.py tries to set up a prompt products database when it is run. This behavior is fine for Alert Production runs against an SQLite database on a single-node machine, but is expected to cause problems when moving to either a more formal (e.g., centrally managed) database system or a more complex computing environment. Therefore, DM-16606 adds a new script, make_ppdb.py, that creates a database for running ap_pipe.py in development environments. The intent is that, with an external script, ap_pipe.py can simply assume the database exists without worrying about whether it should (or can) set it up itself.
The database creation script has been tested and reviewed, but removing the old code from ApPipeTask is a breaking change – users who previously could run ap_pipe.py with no preparation must ensure there's an available prompt products database*. We could have a deprecation period in which users are encouraged to call make_ppdb.py but ap_pipe.py also tries to set up the database (safe to try for SQLite), but that would limit our options for deploying the AP pipeline until the old code is finally removed.
Since I understand that ap_pipe.py is starting to be used by other LSST developers, but hope the number of affected users is still low, I propose to not have a deprecation period and make ap_pipe.py require a pre-existing database as soon as DM-16606 is merged. Is there anyone who would be seriously affected by such a sudden change?
[*Users of ap_verify.py don't need to do anything, only those who call ap_pipe.py directly.]
I'm OK with this; I don't think the original code is widely used, and the new flow seems better for the long term.