Business Context
An automated nightly process kicks off several model scoring jobs in parallel to score different customer segments. Each job runs in its own CAS session for isolation. A master script needs to poll all running job sessions to provide a consolidated status report and trigger downstream actions only when all scoring is complete.
Data Preparation
Create a large dataset and a dummy scoring model to simulate a long-running scoring task.
Copied!
| 1 | DATA casuser.customer_segments; |
| 2 | DO segment_id = 1 to 3; |
| 3 | DO customer_id = 1 to 200000; |
| 4 | var1 = ranuni(456); |
| 5 | var2 = rannor(456); |
| 6 | var3 = ranbin(456, 10, 0.5); |
| 7 | OUTPUT; |
| 8 | END; |
| 9 | END; |
| 10 | RUN; |
| 11 | |
| 12 | PROC CAS; |
| 13 | LOADACTIONSET 'aStore'; |
| 14 | /* Create a dummy model store to simulate scoring */ |
| 15 | SOURCE myModel; |
| 16 | dcl double p_var1; |
| 17 | p_var1=0; |
| 18 | method score(double var1, double var2, double var3, double p_var1); |
| 19 | p_var1 = var1 * 0.5 + var2 * 0.2 + var3 * 0.3; |
| 20 | END; |
| 21 | ENDSOURCE; |
| 22 | ACTION aStore.compile / code=myModel, name='dummy_scorer', caslib='casuser'; |
| 23 | RUN; |
Étapes de réalisation
1
Launch three separate, named sessions and start an asynchronous scoring job in each.
Copied!
| 1 | PROC CAS; |
| 2 | SESSION casauto name='ScoringJob_Seg1'; |
| 3 | SESSION casauto name='ScoringJob_Seg2'; |
| 4 | SESSION casauto name='ScoringJob_Seg3'; |
| 5 | |
| 6 | cas ScoringJob_Seg1; |
| 7 | ACTION aStore.score / TABLE={name='customer_segments', where='segment_id=1'}, rstore={name='dummy_scorer', caslib='casuser'}, out={name='scored_seg1', caslib='casuser', replace=true}, async='job_seg1'; |
| 8 | |
| 9 | cas ScoringJob_Seg2; |
| 10 | ACTION aStore.score / TABLE={name='customer_segments', where='segment_id=2'}, rstore={name='dummy_scorer', caslib='casuser'}, out={name='scored_seg2', caslib='casuser', replace=true}, async='job_seg2'; |
| 11 | |
| 12 | cas ScoringJob_Seg3; |
| 13 | ACTION aStore.score / TABLE={name='customer_segments', where='segment_id=3'}, rstore={name='dummy_scorer', caslib='casuser'}, out={name='scored_seg3', caslib='casuser', replace=true}, async='job_seg3'; |
| 14 | RUN; |
2
From the main session, loop through all sessions prefixed with 'ScoringJob_' and check their status.
Copied!
| 1 | PROC CAS; |
| 2 | SESSION; |
| 3 | ACTION SESSION.listSessions RESULT=r; |
| 4 | DO row over r.SESSION; |
| 5 | IF starts(row['Session Name'], 'ScoringJob_') THEN DO; |
| 6 | PRINT '--- Checking Status for Session: ' || row['Session Name'] || ' (' || row['Session UUID'] || ') ---'; |
| 7 | ACTION SESSION.actionstatus / uuid=row['Session UUID']; |
| 8 | END; |
| 9 | END; |
| 10 | RUN; |
3
Clean up all scoring sessions.
Copied!
| 1 | PROC CAS; |
| 2 | SESSION ScoringJob_Seg1 terminate; |
| 3 | SESSION ScoringJob_Seg2 terminate; |
| 4 | SESSION ScoringJob_Seg3 terminate; |
| 5 | RUN; |
Expected Result
The log will show three distinct blocks of output, one for each scoring session. Each block will contain a status table for that session's 'aStore.score' action. This demonstrates the action's effectiveness in a concurrent, multi-session monitoring scenario, which is typical for batch orchestration.