AWR architecture in a PDB CDB environment

In the previous post we discussed how data sharing works across the PDBs. In this post we will understand the AWR framework works in the multitenant environment. Note that Oracle has made a lot changes between 12.1 and 12.2 and this blog focusses on the 12.2 architecture. Since Oracle 12.2, we can have two different AWR snapshots.

1. AWR snapshot taken at the CDB level. The underlying tables such as WRH$_SQLSTAT will contain data related to root and all of the PDBs.

2. AWR snapshot taken at the PDB level. The underlying tables such as WRH$_SQLSTAT will contain only the current PDBs data.

AWR snapshot at PDB level is disabled by default. To enable it, we need to modify the parameter AWR_PDB_AUTO_FLUSH_ENABLED to TRUE. This parameter is modifiable at PDB level. Since a PDB level and a CDB level snapshot are two different snapshots - it is possible to configure different retention and frequency settings for them. Since we have two different snapshots, this implies that some of the data is redundant, it is stored both in the PDB and the CDB. We will see it shortly.

Underlying tables such as WRH$_SQLSTAT or WRH$_SNAPSHOT are Metadata Linked tables. This means that the dictionary is in the root container but each PDB has its own private data segment. You can consider these as separate tables in each of the PDBs.

Now Oracle exposes this table in the form of views. In earlier versions we had  the DBA_HIST_* views such as DBA_HIST_SQLSTAT or DBA_HIST_SNAPSHOT. But from 12.2 these tables are exposed in different ways.

As you can see, underlying table WRH$_SQLSTAT is exposed via three different views.

AWR_ROOT_SQLSTAT - This is a data link view. If you select from this view you get the data from the CDB's AWR repository.

AWR_PDB_SQLSTAT - This is a metadata link view. If you select from this view you get the data from the current PDB

AWR_CDB_SQLSTAT - This is an extended data link view. If you select from this view you get the data from both the current PDB and the CDB.

DBA_HIST_SQLSTAT is just a view that points to AWR_CDB_SQLSTAT

This is just one example, all other AWR related tables are exposed via similar views.
If we query AWR_ROOT_SQLSTAT, being a data link view, it picks up data from the root container. Remember that CDB's tables contain data from all the PDBs. And if you read this view from a PDB, Oracle is intelligent enough to automatically apply the filter on CON_ID .

SQL> alter session set container=PDB1 ;

SQL> explain plan for
           select * from AWR_ROOT_SQLSTAT ;

Plan hash value: 2631641967

-------------------------------------------------------------------------------------
| Id  | Operation        | Name             | Rows  | Bytes | Cost (%CPU)| Time     |
-------------------------------------------------------------------------------------

|   0 | SELECT STATEMENT |                  |   100 |   209K|     1 (100)| 00:00:01 |
|*  1 |  DATA LINK FULL  | AWR_ROOT_SQLSTAT |   100 |   209K|     1 (100)| 00:00:01 |
-------------------------------------------------------------------------------------

Predicate Information (identified by operation id):
---------------------------------------------------

   1 - filter("CON_ID"=0 OR "CON_ID"=3)

As you can see, since I queried this view from a PDB, Oracle has applied the CON_ID filter.
Therefore, AWR_ROOT_SQLSTAT gets the data from the CDB's WRH$_SQLSTAT after applying the con_id filter. It brings the PDB's data along with the CDB's data itself.

And when we query AWR_CDB_SQLSTAT -since this is an extended data link view it queries the data from the CDB and the current PDB. And since CDB snapshots has the PDB data, you will see duplicate rows here. A SQL will show up twice in this view. One coming from CDB snapshots and other coming from the PDB snapshot. ( This is why it isn't a good idea to collect a PDB level snapshot). If you have a customized script on DBA_HIST_SQLSTAT you may have to modify the script and query AWR_PDB_SQLSTAT instead.

Now that we have covered how the AWR data is stored internally, let us look at the options we have to generate the AWR report.

From a PDB, if you run awrrpt, Oracle asks you to choose the location of AWR data
We get to choose the location of AWR data.  This is because, as explained above, we have different AWR repository, one in the CDB and others in each of the PDBs.
If you choose AWR_ROOT - Oracle then queries AWR_ROOT_* views. Pulls the data from the CDB after applying the con_id filters.

If you choose AWR_PDB - Oracle then queries AWR_PDB_* views. Pulls the data from the current PDBs AWR data. This option is available only when PDB level snapshots are available.

And if you run an AWR report from CDB, you don't get to choose the location. Oracle picks up the data from the root itself using the AWR_PDB_* views.  AWR from CDB is a consolidated report. It displays the information about the each of the PDBs. Since this is a mutitenant architecture we have just one instance, some information such as background wait events are displayed only in the CDB level AWR report.

Conclusion :

PDB level snapshot is clearly an overhead. And this causes same data to be stored in two different places. Probably this is the reason why Oracle chose to disable this by default.
My recommendation would be to keep it disabled. CDB level AWR snaps are enough to get us the diagnostic data. And we can generate the report from either the CDB or the PDB based on what you need to investigate.