[Quality team did an extensive review into Postgres CPU usage](https://gitlab.com/gitlab-org/quality/performance/-/issues/293). During that review it was found that the CPU usage for Postgres look to vary notably with odd increases seen ranging between 10% - 20% higher than the average. As part of the review we did a [long collection of metrics with our day to day performance test runs](https://gitlab.com/gitlab-org/quality/performance/-/issues/293#note_379805918) where this variance was found. For example: GPT test run on 2020-07-28 against 13.3.0-pre `9cda1414a43`:  GPT test run on 2020-07-29 against 13.3.0-pre `e650a4d4c5f`:  Note that the takeaway from the above is the general increase across the board rather than the individual 100% spikes, which are known issues against specific endpoints. The parameters of this testing was as follows: * Environment: [10k Reference Architecture](https://docs.gitlab.com/ee/administration/reference_architectures/10k_users.html) * Postgres Specs - 4 vCPU, 15GB Memory * Upgraded to latest nightly omnibus package before each test run * Later runs were done with `vacuumdb` run manually during upgrade as the environments are only spun up for this testing. * Performance Tests: Standard [GPT](https://gitlab.com/gitlab-org/quality/performance) at 200 RPS (API), 20 RPS (Git and Web) Several areas were investigated as part of the review. Notably comparison to previous versions and if we needed to run `vacuumdb` manually as the environments are typically only spun up for the tests meaning the automated vacuums may be getting missed. All of these additional areas were inconclusive. For our review we were happy that the currently recommended specs look to be correct but this notable variance in CPU was concerning so this should be investigated proper as a potential product issue.