VK Cloud logo

Monitoring

When installing an OS, as well as when ordering a new server, the Enable monitoring option is available. It installs the telegraf monitoring agent.

The agent collects Bare Metal server metrics and sends them to the Cloud Monitoring service. This allows you to track CPU load, memory and disk space usage, network load, and other indicators.

You can view a minimal set of server metrics in the server card in the VK Cloud management console or create your own dashboard with charts.

Bare Metal server metrics

The agent collects operating system metrics every 10 seconds and sends them to the mcs/baremetal namespace.

Bare Metal service labels in Cloud Monitoring:

  • service_id — has the value BMaaS.
  • group_id — has the value Server.
  • stream_id — the server ID, which you can find out in the server card.

CPU

All metrics show the current value summed across all CPU cores.

Label in Cloud Monitoring: cpu=cpu-total.

Metric

Description

cpu_usage_user

The share of time during which the CPU processes user processes.

Used to estimate the load created by user applications.

Unit: %

cpu_usage_system

The share of time during which the CPU processes system processes at the OS kernel level.

Allows you to detect anomalies in system operation, for example, frequent system calls or intensive disk and network activity.

Unit: %

cpu_usage_iowait

The share of time during which the CPU is idle waiting for I/O operations to complete.

Used to identify delays in the storage subsystem. The causes may vary: the drive cannot cope with the load; processes compete for I/O; a driver or low-level software is malfunctioning, or the data bus or the drive itself is faulty.

Unit: %

cpu_usage_irq

The share of time during which the CPU handles hardware interrupts.

Allows you to detect high load when handling requests from hardware, for example, from network adapters under heavy traffic.

Unit: %

cpu_usage_guest

The share of time during which the CPU executes guest OS code.

Allows you to estimate resource utilization by virtual machines if a hypervisor is used on the server.

Unit: %

Memory

All metrics show the current value. Cloud Monitoring labels are not used.

Metric

Description

mem_used_percent

The share of used RAM relative to the total amount.

Used to estimate memory consumption.

Unit: %

mem_used

The amount of used RAM.

Used to estimate memory consumption in absolute values.

Unit: bytes

mem_available_percent

The share of available memory relative to the total amount.

Allows you to estimate the RAM available for new user applications.

Unit: %

mem_available

The amount of memory available to applications without using the swap file.

Allows you to estimate the RAM available for new user applications in absolute values.

Unit: bytes

mem_cached

The amount of memory occupied by the file system cache.

Allows you to estimate the amount of RAM used by the OS to speed up disk operations.

Unit: bytes

mem_dirty

The amount of data in memory queued for writing.

Allows you to identify delays in the storage subsystem if the value is high or grows quickly together with cpu_usage_iowait. Possible causes are described for the cpu_usage_iowait metric.

Unit: bytes

mem_free

The amount of memory that is not used in any way.

Allows you to distinguish truly free memory from other memory types.

Unit: bytes

File systems

The current value is collected for each file system.

Cloud Monitoring labels:

  • device — the device name without the /dev/ prefix;
  • path — the mount point;
  • fstype — the file system type.

The system file systems tmpfs, devtmpfs, devfs, iso9660, overlay, aufs, squashfs are not taken into account.

Metric

Description

disk_used_percent

File system usage.

Used to track drive space utilization.

Unit: %

disk_free

The amount of free space in the file system.

Used to estimate the available drive capacity in absolute values.

Unit: bytes

disk_inodes_free

The number of free index descriptors (inodes) in the file system.

Allows you to detect a shortage of index descriptors (inodes) for storing metadata. Their number is limited, and if there are many small files, inodes may run out before disk space does, which makes it impossible to create new objects.

Unit: pieces

Disk I/O

A separate value is collected for each disk. All metrics show the counter value since the OS boot.

Cloud Monitoring uses the name label — the device name without the /dev/ prefix.

Metric

Description

diskio_reads

The number of completed read operations.

Used to estimate the intensity of read operations (IOPS).

Unit: pieces

diskio_writes

The number of completed write operations.

Used to estimate the intensity of write operations (IOPS).

Unit: pieces

diskio_read_bytes

The amount of data read.

Used to estimate the disk read speed by comparing intermediate values.

Unit: bytes

diskio_write_bytes

The amount of data written.

Used to estimate the disk write speed by comparing intermediate values.

Unit: bytes

diskio_read_time

The total time spent on read operations.

Allows you to identify an increase in read latency by comparing intermediate values.

Unit: milliseconds

diskio_write_time

The total time spent on write operations.

Allows you to identify an increase in write latency by comparing intermediate values.

Unit: milliseconds

diskio_io_time

The time during which the disk was busy with I/O operations.

Allows you to identify an increase in the total time spent on I/O operations by comparing intermediate values.

Unit: milliseconds

diskio_weighted_io_time

The disk busy time weighted by the number of requests in the queue.

Allows you to identify disk overload: the value grows faster than diskio_io_time if requests are queued.

Unit: milliseconds

Network

A separate value is collected for each network interface. Except for netstat_tcp_established, all metrics show the counter value since the OS boot. For netstat_tcp_established, the current value is shown.

Cloud Monitoring uses the interface label — the network interface name used at the Linux kernel level.

Metric

Description

net_bytes_sent

The amount of data sent.

Used to estimate outgoing traffic.

Unit: bytes

net_bytes_recv

The amount of data received.

Used to estimate incoming traffic.

Unit: bytes

net_packets_sent

The number of packets sent.

Used to estimate the intensity of outgoing traffic in packets.

Unit: pieces

net_packets_recv

The number of packets received.

Used to estimate the intensity of incoming traffic in packets.

Unit: pieces

net_err_in

The number of errors when receiving packets.

Allows you to identify problems with network equipment or the communication channel by an increase in the count.

Unit: pieces

net_err_out

The number of errors when sending packets.

Allows you to identify problems with network equipment or the communication channel by an increase in the count.

Unit: pieces

net_drop_in

The number of dropped incoming packets.

Allows you to identify overload of the network interface or the OS.

Unit: pieces

net_drop_out

The number of dropped outgoing packets.

Allows you to identify overload of the network interface or the OS.

Unit: pieces

netstat_tcp_established

The number of established TCP connections.

Used to estimate the number of active connections to the server.

Unit: pieces

GPU

For servers with a GPU, a separate value is collected for each device. All metrics show the current value requested using the nvidia-smi utility.

Cloud Monitoring uses the following labels:

  • index — the GPU sequence number on the server, starting from 0;
  • uuid — the unique GPU ID;
  • name — the GPU model.

Metric

Description

nvidia_smi_utilization_gpu

GPU compute core utilization.

Used to estimate the GPU load.

Unit: %

nvidia_smi_utilization_memory

GPU memory bus utilization.

Allows you to identify performance limitations caused by VRAM access speed.

Unit: %

nvidia_smi_memory_used

The amount of used GPU memory.

Used to track VRAM utilization.

Unit: MiB

nvidia_smi_memory_free

The amount of free GPU memory.

Used to track free VRAM capacity.

Unit: MiB

nvidia_smi_memory_total

The total amount of GPU memory.

Allows you to estimate the total VRAM capacity and compare it with the nvidia_smi_memory_used and nvidia_smi_memory_free values.

Unit: MiB

nvidia_smi_temperature_gpu

GPU temperature.

Used to track temperature in order to prevent frequency throttling.

Unit: °C

nvidia_smi_power_draw

The power consumed by the GPU.

Used to estimate power consumption.

Unit: W

nvidia_smi_clocks_current_graphics

The current frequency of the graphics processor.

Allows you to track frequency throttling caused by overheating or power limits.

Unit: MHz

nvidia_smi_clocks_current_memory

The current GPU memory frequency.

Allows you to track memory frequency reduction caused by overheating or power limits.

Unit: MHz

nvidia_smi_fan_speed

The GPU fan speed as a percentage of the maximum.

Used to check the operation of the cooling system.

Unit: %

nvidia_smi_pcie_link_gen_current

The current generation of the PCIe interface through which the GPU is connected.

Allows you to analyze GPU performance taking the interface generation into account.

Unit: generation number

nvidia_smi_pcie_link_width_current

The current number of PCIe lanes through which the GPU is connected.

Allows you to analyze GPU performance taking the number of lanes into account.

Unit: number of lanes

nvidia_smi_retired_pages_double_bit

The number of GPU memory pages retired due to uncorrectable ECC errors.

Allows you to identify VRAM faults.

Unit: pieces

nvidia_smi_remapped_rows_uncorrectable

The number of GPU memory rows remapped due to uncorrectable errors.

Allows you to identify VRAM faults.

Unit: pieces

Standard dashboard charts

Chart

Description

RAM Used

Allows you to estimate memory consumption.

Metrics: mem_used_percent.

Unit: %

Current IOWait

Allows you to identify delays in the storage subsystem. Possible causes are described for the cpu_usage_iowait metric.

Metrics: cpu_usage_iowait.

Unit: %

Current CPU

Allows you to estimate the overall CPU load.

Metrics: the sum of cpu_usage_user, cpu_usage_system, cpu_usage_iowait, cpu_usage_irq, cpu_usage_guest.

Unit: %

Disk Used

Allows you to track file system space utilization.

Metrics: disk_used_percent (the maximum value among all file systems).

Unit: %

CPU

Allows you to track how the application load and I/O wait change over time.

Metrics: cpu_usage_user (the User series), cpu_usage_iowait (the IO wait series).

Unit: %

RAM

Allows you to track how RAM usage changes over time.

Metrics: mem_used_percent (the Used series).

Unit: %

Disk IO

Allows you to track how the disk read and write speed changes over time.

Metrics: diskio_read_bytes (the Reads series), diskio_write_bytes (the Writes series), as a rate of change (rate), summed across all disks.

Unit: bytes/s

Disk used

Allows you to track how the usage of each file system changes over time.

Metrics: disk_used_percent, one series for each file system.

Unit: %

Network

Allows you to track how the speed of sending and receiving data over the network changes over time.

Metrics: net_bytes_sent (the Sent series), net_bytes_recv (the Received series), as a rate of change (rate), summed across all network interfaces.

Unit: bytes/s

Was this article helpful?