Update GV$PROCESS query #14143
Update GV$PROCESS query #14143
Conversation
Update GV$PROCESS query to better handle multiple records with the same program value
ghost
added
the
|
Here is an example of the improvement:
|
|
👋 hey @jake-condello looks like the style tests are failing Creating environment: lint
Checking dependencies
Syncing dependencies
cmd [1] | flake8 --config=../.flake8 .
./datadog_checks/oracle/queries.py:8:121: E501 line too long (143 > 120 characters)
Failed!you can run this these locally (assuming your ddev is setup) by doing |
| @@ -5,7 +5,7 @@ | |||
|
|
|||
| ProcessMetrics = { | |||
| 'name': 'process', | |||
| 'query': 'SELECT PROGRAM, PGA_USED_MEM, PGA_ALLOC_MEM, PGA_FREEABLE_MEM, PGA_MAX_MEM FROM GV$PROCESS', | |||
| 'query': 'SELECT PROGRAM, SUM(PGA_USED_MEM), SUM(PGA_ALLOC_MEM), SUM(PGA_FREEABLE_MEM), SUM(PGA_MAX_MEM) FROM GV$PROCESS GROUP BY PROGRAM', | |||
There was a problem hiding this comment.
What's the purpose of summing up pga_max_mem? pga_max_mem is a high usage watermark for every process. But since every process generally reached its max usage at a different time, their sum will be greater than the total memory ever allocated.
There was a problem hiding this comment.
That's a good point; the sum of all processes' maximums isn't the most useful.
For some context, the main goal for our company is to have a good idea of the PGA state (used, total, etc.). Our current dashboards are, for example, summing up oracle.process.pga_used_memory and reporting that as amount of PGA used. Based on my research that is incorrect because only one value is reported per program when multiple exist:
select program, count(*) from GV$PROCESS group by program order by 2 desc;
oracle@cs12dbnrduva01 356
oracle@cs12dbnrduva01 (SCMN) 2
My first thought was to sum them all, grouped by program so that there is no information missing. Do you have any thoughts on how else we can bring in all that missing information?
There was a problem hiding this comment.
Maybe we can add PID column to query so that the entries with duplicate program names are not overridden. For example:
SELECT PID, PROGRAM, PGA_USED_MEM, PGA_ALLOC_MEM, PGA_FREEABLE_MEM, PGA_MAX_MEM FROM GV$PROCESS
There was a problem hiding this comment.
Yes, program can't be the key because generally a single program can have multiple sessions/processes.
Summing memory allocations and grouping by program solves this - except for pga_max_mem where summing doesn't produce a meaningful value.
An alternative solution, adding PID as a unique identifier seems correct to me. The aggregations would need to be performed outside of database in that case. It would require a thorough check that adding PID doesn't break something else.
There was a problem hiding this comment.
My latest commit includes the original query with the pid column added. ddev test oracle passes on my machine.
Previously existing monitors seem to be functioning with correct values
And all metrics appear to be reporting properly
There was a problem hiding this comment.
@nenadnoveljic Is there anything else that should be done before this can be merged?
Update GV$PROCESS query to better handle multiple records with the same program value
What does this PR do?
This PR fixes the GV$PROCESS query so that the program value in each record returned is unique.
Motivation
The current query returns multiple records with the same program value. And since program is the only tag, only one of the values is reported, and all the other values are omitted within DataDog. Based on my testing, only the values associated with the last record of a given program name is reported. This leads to incorrect calculations of PGA information in the database.
Additional Notes
Review checklist (to be filled by reviewers)
changelog/andintegration/labels attachedqa/skip-qalabel.