Dashboard: Compute Checks
Thrifty Compute Benchmark
Thrifty developers eliminate their unused and under-utilized compute instances. This benchmark focuses on finding resources that have not been restarted recently, have old snapshots, have high disk IOPS and have large, unused or inactive disks.
Variables
Variable | Description | Default |
---|---|---|
compute_disk_avg_read_write_ops_high | The number of average read/write ops required for disks to be considered frequently used. This value should be higher than compute_disk_avg_read_write_ops_low . | 500 |
compute_disk_avg_read_write_ops_low | The number of average read/write ops required for disks to be considered infrequently used. This value should be lower than compute_disk_avg_read_write_ops_high . | 100 |
compute_disk_max_iops | The maximum IOPS allowed for disks. | 2000 IOPS |
compute_disk_max_size_gb | The maximum size (GB) allowed for disks. | 100 GB |
compute_running_vm_age_max_days | The maximum number of days virtual machines are allowed to run. | 90 days |
compute_snapshot_age_max_days | The maximum number of days snapshots can be retained. | 90 days |
compute_vm_avg_cpu_utilization_high | The average CPU utilization required for virtual machines to be considered frequently used. This value should be higher than compute_vm_avg_cpu_utilization_low . | 35% |
compute_vm_avg_cpu_utilization_low | The average CPU utilization required for virtual machines to be considered infrequently used. This value should be lower than compute_vm_avg_cpu_utilization_high . | 20% |
Usage
Install the mod:
mkdir dashboardscd dashboardspowerpipe mod initpowerpipe mod install github.com/turbot/steampipe-mod-azure-thrifty
Start the Powerpipe server:
steampipe service startpowerpipe server
Open http://localhost:9033 in your browser and select Compute Checks dashboard.
You could also snapshot and share results via Turbot Pipes:
powerpipe loginpowerpipe benchmark run azure_thrifty.benchmark.compute --share
Benchmark
This dashboard is automatically generated from the following benchmark:
benchmark.compute