Content-type: text/html Manpage of CBLS.REMOTE

CBLS.REMOTE

Section: Maintenance Commands (8)
Updated: SkyForm AIP Version 10.25.0 - April 2025
Index Return to Main Contents

NAME

cbls.remote - Remote host load sensor

SYNOPSIS

CB_SERVERDIR/cbls.remote

DESCRIPTION

cbls.remote is a customized remote host load sensor in AIP. It is executed by the AIP load server cbls periodically to obtain the information and load of remote servers.

cbls.remote must reside in the directory of CB_SERVERDIR, e.g. /opt/skyformai/sbin and must be executable.

CBLS.REMOTE CODE LOGIC

cbls.remote code logic should get required data from all remote hosts. Unlike RESS, it should not be an infinite loop, i.e. the master cbls calls it perodically. It is only called by the master cbls. It should output YAML data, then exit.

Example of cbls.remote output:

---
remote_hosts:
- name: node0013

  model: x6_5424_IntelRCoreTMi57200UCPU250GHz

  type: x86_64

  status: ok

  cpuf: 54.2

  sockets: 4

  cores_per_socket: 1

  cpus: 4

  maxmem: 1980

  maxswap: 2047

  gpus:

  - id: 0

    model: GeForce RTX 2080 Ti

    totalmem: 11019

  - id: 1

    model: GeForce RTX 2080 Ti

    totalmem: 11019

  ut: 0.23

  r1m: 1.23

  mem: 1311.0

  up: 42507
...

The command 'cbls -y' provides the sample output of the remote host data.

OUTPUT FORMAT

The output of cbls.remote should follow YAML standard. The first line is always: 'remote_hosts:'. Then an array of data. Each element of the array represents a remote host configured in cb.yaml.

For each remote host, the elements are:

name:

The host name. The host name must be configured in cb.yaml as a remote host, or the host data from cbls.remote is ignored. This field is mandatory.

model:

The CPU model. If the field is missing, the remote host will be marked with the same CPU model as the master host. CPU model can be any string. This field is optional.

type:

The host type, for example: x86_64. If the field is missing, the remote host will be marked with the same host model as the master host. This field is optional.

status:

If the host is in good condition to run job, the value should be "ok". If not, put any other string. This field is mandatory.

cpuf:

A numeric value represents the performance of the host. The typical value in Linux is the BogoMIPS in the output of 'lscpu' divided by 100. If missing, it will take the same value as the master host. This field is optional.

sockets:

The number of CPU sockets on the host. This field is optional.

cores_per_socket:

The number of CPU cores per CPU socket. This field is optional.

cpus:

The total number of logical CPU cores. This field is optional. The default value is 1.

maxmem:

Maximum memory in MB installed on the host. This field is mandatory.

maxswap:

Maximum swap space in MB configured on the host. This field is optional.

gpus:

An array of GPU data. This field is optional.

id:: The GPU ID. Typically 0, 1, 2 etc. This field is mandatory if the gpus section exists.
model:: The GPU model. For example: Tesla P100. This field is optional.
totalmem:: The total installed GPU memory for the GPU in MB. This field is optional.

ut:

The CPU utilization exponentially averaged over the last minute, between 0 and 1. This field is optional.

r15s:

The 15-second exponentially averaged CPU run queue length. This field is optional.

r1m:

The 1-minute exponentially averaged CPU run queue length. This field is optional.

r15m:

The 15-minute exponentially averaged CPU run queue length. This field is optional.

pg:

The memory paging rate exponentially averaged over the last minute, in pages per second. This field is optional.

io:

The network I/O rate exponentially averaged over the last minute, in KB per second. This field is optional.

up:

Host uptime in minutes. This field is optional.

swap:

The amount of swap space available, in megabytes. This field is optional.

mem:

The amount of available memory, in megabytes. This field is optional.

tmp:

The amount of free space in /tmp, in megabytes. This field is optional.

This document was created by man2html, using the manual pages.
Time: 18:57:47 GMT, April 23, 2025