Querying for &attributes=hardware.cpu_sockets is 0.5 seconds faster than num_cpu, WAT

[jrafanie]

joerafaniello@Joes-MBP-2 manageiq % bundle exec miqperf benchmark -ac 5 -d "/api/vms?expand=resources&attributes=name,hardware.cpu_sockets"
D, [2020-07-24T14:55:14.826645 #30092] DEBUG -- : --> logging in...
D, [2020-07-24T14:55:15.048047 #30092] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,hardware.cpu_sockets
D, [2020-07-24T14:55:16.068500 #30092] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,hardware.cpu_sockets
D, [2020-07-24T14:55:17.008414 #30092] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,hardware.cpu_sockets
D, [2020-07-24T14:55:18.018111 #30092] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,hardware.cpu_sockets
D, [2020-07-24T14:55:18.906843 #30092] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,hardware.cpu_sockets
joerafaniello@Joes-MBP-2 manageiq % bundle exec miqperf report -l
/api/vms
|   ms | queries | query (ms) | rows |
| ---: |    ---: |       ---: | ---: |
|  952 |     563 |      405.8 | 1112 |
|  875 |     563 |      392.4 | 1112 |
|  884 |     563 |      393.9 | 1112 |
|  824 |     563 |      371.7 | 1112 |
|  890 |     563 |      392.8 | 1112 |

joerafaniello@Joes-MBP-2 manageiq % bundle exec miqperf benchmark -ac 5 -d "/api/vms?expand=resources&attributes=name,num_cpu"
D, [2020-07-24T14:55:46.390251 #30111] DEBUG -- : --> logging in...
D, [2020-07-24T14:55:46.600145 #30111] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,num_cpu
D, [2020-07-24T14:55:48.762894 #30111] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,num_cpu
D, [2020-07-24T14:55:50.871009 #30111] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,num_cpu
D, [2020-07-24T14:55:52.901907 #30111] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,num_cpu
D, [2020-07-24T14:55:54.991955 #30111] DEBUG -- : --> making GET request: /api/vms?expand=resources&attributes=name,num_cpu
joerafaniello@Joes-MBP-2 manageiq % bundle exec miqperf report -l
/api/vms
|   ms | queries | query (ms) | rows |
| ---: |    ---: |       ---: | ---: |
| 2010 |    1116 |      983.7 | 1665 |
| 1938 |    1116 |      955.4 | 1665 |
| 1873 |    1116 |      934.4 | 1665 |
| 1970 |    1116 |      956.0 | 1665 |
| 2017 |    1116 |      996.3 | 1665 |

[kbrock]

I guess it would be nice if we knew how to swap between virtual attributes and delegation
a lot of that could be based upon creating the extra object

but I think of how I would try and solve this and I don't want to support my imagined code

[kbrock]

I wonder if the performance hit is due to the fact that virtual attributes subqueries are run for every row, even those that are not brought back.