This chapter describes using Oracle Scheduler to schedule jobs in a multitenant container database (CDB). This chapter makes the following assumptions:
You understand how to configure and manage a CDB. See Part VI, "Managing a Multitenant Environment" for an overview and related information.
You understand how to use Oracle Scheduler to schedule jobs in a non-CDB. See Chapter 28, Chapter 29, and Chapter 30 for information.
This chapter contains the following topics:
Most scheduler calls work exactly the same way as they did in non-CDBs, with the exception of two scheduler global attributes. To limit job slaves, set the value of the job_queue_processes initialization parameter.
For all other global attribute settings, you must be at the pluggable database (PDB) level only. For example, if you set the EMAIL_SENDER attribute in the root database, it applies to the jobs that run in the root, not the jobs running in a specific PDB. If you want to pick a new EMAIL_SENDER for a PDB, then you must set the global attribute in that PDB.
The major CDB-related changes are to the job coordinator process.
In a non-CDB, the coordinator looks at all jobs that are ready to run, picks a subset of them to run, and assigns them to job slaves. It also opens and closes windows, which changes the resource plan in effect for the database.
That is essentially what happens inside a CDB except for the following:
Jobs are selected from all PDBs
The coordinator looks at the root database and all the child PDBs and selects jobs based on the job priority, the job scheduled start time, and the availability of resources to run the job. The latter criterion depends on the consumer group of the job and the resource plan currently in effect. The coordinator makes no attempt to be fair to every PDB. The only way to ensure that jobs from a PDB are not starved is to allocate enough resources to it.
Windows are open in the PDB and root database levels
In a non-CDB, only one window can be open at any given time. In a CDB, there are two levels of windows. At the PDB level, windows can be used to set resource plans that allocate resources among consumer groups belonging to that PDB. At the root database level, windows can be used to allocate resources to various different PDBs. Therefore, at any time, there can be a window open in the root database and one in each PDB.
Job slave switches to the specific PDB it belongs to
The job slaves are essentially the same as in a non-CDB, except that when a slave executes a job, it switches to the PDB that the job belongs to and then executes it. The rest of the code is essentially unchanged.
You can create a job using DBMS_JOB within a PDB, and it will work as before. However, DBMS_JOB has been desupported and using it is not recommended.
For the scheduler, the coordinator now selects jobs to run from every single PDB and not just a single database as was the case before. Also, for the scheduler, the slave will switch into a PDB before executing a job; otherwise, the code is essentially unchanged.
See Also:
Appendix A, "Support for DBMS_JOB"If a PDB is closed with the immediate option, then the coordinator terminates jobs running in the PDB, and the jobs must be recovered before they can run again. In an Oracle RAC database, the coordinator can, in most cases, recover the jobs on another instance where that PDB is open. So, if the coordinator on the first instance can find another instance where the PDB is still open, it moves the jobs there. In certain cases, moving the jobs to another instance may not be possible. For example, if the PDB in question is not open anywhere else, the jobs cannot be moved. Also, moving a job to another instance is not possible when the job has the INSTANCE_ID attribute set. In this case the job cannot run until the PDB on that instance is open again.
In a non-Oracle RAC case, the question of moving jobs does not arise. Terminated jobs can only be recovered after the PDB is opened again.
With the CDB, changes have been made to existing views and new views have been added. See Oracle Database Reference for details.
V$ and GV$ views now have an additional column (CON_ID) which identifies a container whose data a given CDB_* row represents. In non-CDBs, this column is NULL.
There are CDB_* views corresponding to all Scheduler DBA_* views.
In a PDB, these views only show objects visible through a corresponding DBA_* view, but all objects can be viewed by the root database. The CDB_* view contains all columns found in a given DBA_* view and the column (CON_ID). In non-CDBs, this column is NULL.