Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

feat(intel_powerstat): Add uncore frequency metrics #11254

Merged
merged 1 commit into from
Jun 6, 2022
Merged

feat(intel_powerstat): Add uncore frequency metrics #11254

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
99 changes: 66 additions & 33 deletions plugins/inputs/intel_powerstat/README.md
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,10 +1,13 @@
# Intel PowerStat Input Plugin

This input plugin monitors power statistics on Intel-based platforms and assumes presence of Linux based OS.
This input plugin monitors power statistics on Intel-based platforms and assumes
presence of Linux based OS.

Main use cases are power saving and workload migration. Telemetry frameworks allow users to monitor critical platform level metrics.
Key source of platform telemetry is power domain that is beneficial for MANO/Monitoring&Analytics systems
to take preventive/corrective actions based on platform busyness, CPU temperature, actual CPU utilization and power statistics.
Main use cases are power saving and workload migration. Telemetry frameworks
allow users to monitor critical platform level metrics. Key source of platform
telemetry is power domain that is beneficial for MANO Monitoring&Analytics
systems to take preventive/corrective actions based on platform busyness, CPU
temperature, actual CPU utilization and power statistics.

## Configuration

Expand All @@ -16,7 +19,7 @@ to take preventive/corrective actions based on platform busyness, CPU temperatur
## - Setting this value to an empty array means no package metrics will be collected
## - Finally, a user can specify individual metrics to capture from the supported options list
## Supported options:
## "current_power_consumption", "current_dram_power_consumption", "thermal_design_power", "max_turbo_frequency"
## "current_power_consumption", "current_dram_power_consumption", "thermal_design_power", "max_turbo_frequency", "uncore_frequency"
# package_metrics = ["current_power_consumption", "current_dram_power_consumption", "thermal_design_power"]

## The user can choose which per-CPU metrics are monitored by the plugin in cpu_metrics array.
Expand All @@ -29,7 +32,8 @@ to take preventive/corrective actions based on platform busyness, CPU temperatur

## Example: Configuration with no per-CPU telemetry

This configuration allows getting default processor package specific metrics, no per-CPU metrics are collected:
This configuration allows getting default processor package specific metrics, no
per-CPU metrics are collected:

```toml
[[inputs.intel_powerstat]]
Expand All @@ -38,15 +42,17 @@ This configuration allows getting default processor package specific metrics, no

## Example: Configuration with no per-CPU telemetry - equivalent case

This configuration allows getting default processor package specific metrics, no per-CPU metrics are collected:
This configuration allows getting default processor package specific metrics, no
per-CPU metrics are collected:

```toml
[[inputs.intel_powerstat]]
```

## Example: Configuration for CPU Temperature and CPU Frequency

This configuration allows getting default processor package specific metrics, plus subset of per-CPU metrics (CPU Temperature and CPU Frequency):
This configuration allows getting default processor package specific metrics,
plus subset of per-CPU metrics (CPU Temperature and CPU Frequency):

```toml
[[inputs.intel_powerstat]]
Expand All @@ -55,7 +61,8 @@ This configuration allows getting default processor package specific metrics, pl

## Example: Configuration for CPU Temperature and CPU Frequency without default package metrics

This configuration allows getting only a subset of per-CPU metrics (CPU Temperature and CPU Frequency):
This configuration allows getting only a subset of per-CPU metrics (CPU
Temperature and CPU Frequency):

```toml
[[inputs.intel_powerstat]]
Expand All @@ -65,27 +72,33 @@ This configuration allows getting only a subset of per-CPU metrics (CPU Temperat

## Example: Configuration with all available metrics

This configuration allows getting all processor package specific metrics and all per-CPU metrics:
This configuration allows getting all processor package specific metrics and all
per-CPU metrics:

```toml
[[inputs.intel_powerstat]]
package_metrics = ["current_power_consumption", "current_dram_power_consumption", "thermal_design_power", "max_turbo_frequency"]
package_metrics = ["current_power_consumption", "current_dram_power_consumption", "thermal_design_power", "max_turbo_frequency", "uncore_frequency"]
cpu_metrics = ["cpu_frequency", "cpu_busy_frequency", "cpu_temperature", "cpu_c0_state_residency", "cpu_c1_state_residency", "cpu_c6_state_residency"]
```

## SW Dependencies

Plugin is based on Linux Kernel modules that expose specific metrics over `sysfs` or `devfs` interfaces.
The following dependencies are expected by plugin:
Plugin is based on Linux Kernel modules that expose specific metrics over
`sysfs` or `devfs` interfaces. The following dependencies are expected by
plugin:

- _intel-rapl_ module which exposes Intel Runtime Power Limiting metrics over `sysfs` (`/sys/devices/virtual/powercap/intel-rapl`),
- _msr_ kernel module that provides access to processor model specific registers over `devfs` (`/dev/cpu/cpu%d/msr`),
- _cpufreq_ kernel module - which exposes per-CPU Frequency over `sysfs` (`/sys/devices/system/cpu/cpu%d/cpufreq/scaling_cur_freq`).
- _intel-uncore-frequency_ module exposes Intel uncore frequency metrics over `sysfs` (`/sys/devices/system/cpu/intel_uncore_frequency`),

Minimum kernel version required is 3.13 to satisfy all requirements.
Minimum kernel version required is 3.13 to satisfy most of requirements,
for `uncore_frequency` metrics `intel-uncore-frequency` module is required
(available since kernel 5.6).

Please make sure that kernel modules are loaded and running (cpufreq is integrated in kernel). Modules might have to be manually enabled by using `modprobe`.
Depending on the kernel version, run commands:
Please make sure that kernel modules are loaded and running (cpufreq is
integrated in kernel). Modules might have to be manually enabled by using
`modprobe`. Depending on the kernel version, run commands:

```sh
# kernel 5.x.x:
Expand All @@ -94,13 +107,17 @@ subo modprobe msr
sudo modprobe intel_rapl_common
sudo modprobe intel_rapl_msr

# also for kernel >= 5.6.0
sudo modprobe intel-uncore-frequency

# kernel 4.x.x:
sudo modprobe msr
sudo modprobe intel_rapl
```

**Telegraf with Intel PowerStat plugin enabled may require root access to read model specific registers (MSRs)**
to retrieve data for calculation of most critical per-CPU specific metrics:
**Telegraf with Intel PowerStat plugin enabled may require root access to read
model specific registers (MSRs)** to retrieve data for calculation of most
critical per-CPU specific metrics:

- `cpu_busy_frequency_mhz`
- `cpu_temperature_celsius`
Expand All @@ -111,17 +128,20 @@ to retrieve data for calculation of most critical per-CPU specific metrics:
and to retrieve data for calculation per-package specific metric:

- `max_turbo_frequency_mhz`
- `uncore_frequency_mhz_cur`

To expose other Intel PowerStat metrics root access may or may not be required (depending on OS type or configuration).
To expose other Intel PowerStat metrics root access may or may not be required
(depending on OS type or configuration).

## HW Dependencies

Specific metrics require certain processor features to be present, otherwise Intel PowerStat plugin won't be able to
read them. When using Linux Kernel based OS, user can detect supported processor features reading `/proc/cpuinfo` file.
Specific metrics require certain processor features to be present, otherwise
Intel PowerStat plugin won't be able to read them. When using Linux Kernel based
OS, user can detect supported processor features reading `/proc/cpuinfo` file.
Plugin assumes crucial properties are the same for all CPU cores in the system.
The following processor properties are examined in more detail in this section:
processor _cpu family_, _model_ and _flags_.
The following processor properties are required by the plugin:
processor _cpu family_, _model_ and _flags_. The following processor properties
are required by the plugin:

- Processor _cpu family_ must be Intel (0x6) - since data used by the plugin assumes Intel specific
model specific registers for all features
Expand Down Expand Up @@ -186,9 +206,11 @@ and _powerstat\_core.cpu\_c6\_state\_residency_ metrics:

## Metrics

All metrics collected by Intel PowerStat plugin are collected in fixed intervals.
Metrics that reports processor C-state residency or power are calculated over elapsed intervals.
When starting to measure metrics, plugin skips first iteration of metrics if they are based on deltas with previous value.
All metrics collected by Intel PowerStat plugin are collected in fixed
intervals. Metrics that reports processor C-state residency or power are
calculated over elapsed intervals. When starting to measure metrics, plugin
skips first iteration of metrics if they are based on deltas with previous
value.

**The following measurements are supported by Intel PowerStat plugin:**

Expand Down Expand Up @@ -225,39 +247,50 @@ When starting to measure metrics, plugin skips first iteration of metrics if the
|-----|-------------|
| `package_id` | ID of platform package/socket |
| `active_cores`| Specific tag for `max_turbo_frequency_mhz` metric. The maximum number of activated cores for reachable turbo frequency
| `die`| Specific tag for all `uncore_frequency` metrics. Id of die
| `type`| Specific tag for all `uncore_frequency` metrics. Type of uncore frequency (current or initial)

Measurement powerstat_package metrics are collected per processor package -_package_id_ tag indicates which package metric refers to.

- Available metrics for powerstat_package measurement

| Metric name (field) | Description | Units |
|-----|-------------|-----|
| `thermal_design_power_watts` | Maximum Thermal Design Power (TDP) available for processor package | Watts |
| `thermal_design_power_watts` | Maximum Thermal Design Power (TDP) available for processor package | Watts |
| `current_power_consumption_watts` | Current power consumption of processor package | Watts |
| `current_dram_power_consumption_watts` | Current power consumption of processor package DRAM subsystem | Watts |
| `max_turbo_frequency_mhz`| Maximum reachable turbo frequency for number of cores active | MHz
| `uncore_frequency_limit_mhz_min`| Minimum uncore frequency limit for die in processor package | MHz
| `uncore_frequency_limit_mhz_max`| Maximum uncore frequency limit for die in processor package | MHz
| `uncore_frequency_mhz_cur`| Current uncore frequency for die in processor package. Available only with tag `current`. Since this value is not yet available from `intel-uncore-frequency` module it needs to be accessed via MSR. In case of lack of loaded msr, only `uncore_frequency_limit_mhz_min` and `uncore_frequency_limit_mhz_max` metrics will be collected | MHz

### Known issues

From linux kernel version v5.4.77 with [this kernel change](https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=v5.4.77&id=19f6d91bdad42200aac557a683c17b1f65ee6c94)
resources like `/sys/class/powercap/intel-rapl*/*/energy_uj` are readable only by root for security reasons, so this plugin needs root privileges to work properly.
From linux kernel version v5.4.77 with [this kernel change][19f6d91b] resources
like `/sys/class/powercap/intel-rapl*/*/energy_uj` are readable only by root for
security reasons, so this plugin needs root privileges to work properly.

If such strict security restrictions are not relevant, reading permissions to files in `/sys/devices/virtual/powercap/intel-rapl/`
directory can be manually changed for example with `chmod` command with custom parameters.
For example to give all users permission to all files in `intel-rapl` directory:
If such strict security restrictions are not relevant, reading permissions to
files in `/sys/devices/virtual/powercap/intel-rapl/` directory can be manually
changed for example with `chmod` command with custom parameters. For example to
give all users permission to all files in `intel-rapl` directory:

```bash
sudo chmod -R a+rx /sys/devices/virtual/powercap/intel-rapl/
```

### Example Output
[19f6d91b]: https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux.git/commit/?h=v5.4.77&id=19f6d91bdad42200aac557a683c17b1f65ee6c94

## Example Output

```shell
powerstat_package,host=ubuntu,package_id=0 thermal_design_power_watts=160 1606494744000000000
powerstat_package,host=ubuntu,package_id=0 current_power_consumption_watts=35 1606494744000000000
powerstat_package,host=ubuntu,package_id=0 current_dram_power_consumption_watts=13.94 1606494744000000000
powerstat_package,host=ubuntu,package_id=0,active_cores=0 max_turbo_frequency_mhz=3000i 1606494744000000000
powerstat_package,host=ubuntu,package_id=0,active_cores=1 max_turbo_frequency_mhz=2800i 1606494744000000000
powerstat_package,die=0,host=ubuntu,package_id=0,type=initial uncore_frequency_limit_mhz_min=800,uncore_frequency_limit_mhz_max=2400 1606494744000000000
powerstat_package,die=0,host=ubuntu,package_id=0,type=current uncore_frequency_mhz_cur=800i,uncore_frequency_limit_mhz_min=800,uncore_frequency_limit_mhz_max=2400 1606494744000000000
powerstat_core,core_id=0,cpu_id=0,host=ubuntu,package_id=0 cpu_frequency_mhz=1200.29 1606494744000000000
powerstat_core,core_id=0,cpu_id=0,host=ubuntu,package_id=0 cpu_temperature_celsius=34i 1606494744000000000
powerstat_core,core_id=0,cpu_id=0,host=ubuntu,package_id=0 cpu_c6_state_residency_percent=92.52 1606494744000000000
Expand Down
17 changes: 16 additions & 1 deletion plugins/inputs/intel_powerstat/file_mock_test.go
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

Loading