linux服务器显卡监控脚本

文章目录

• 1.watch_nvidia
• 2.send_msg
• 3.完整脚本nvidia.py
• 4.后台运行

---

• 前期准备：
  pip install pynvml

1.watch_nvidia

#参数：nvidia_ids：显卡id   min_memory：最小可用显存 GB
def watch_nvidia(nvidia_ids,min_memory):
  flag = [1 for i in nvidia_ids]
  for i in nvidia_ids:
    handle = pynvml.nvmlDeviceGetHandleByIndex(i)
    meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
    #遍历每块卡的剩余显存
    print("card {} free memory is {}GB".format(i,meminfo.free * 1.0 /(1024**3)))
    if meminfo.free * 1.0 /(1024**3) > min_memory:
      flag[i-1]=0
    else:
      flag[i-1]=1#统计符合要求的卡的数量
  if 0 in flag:
    free_num = 0
    for i in flag:
      if i == 0:
        free_num += 1
    return free_num#返回符合要求的卡的数量
  else:
    print("no free card!")
    return -1

2.send_msg

#发送邮箱信息
#target_email:接受信息的邮箱，msg：发送的消息
def send_msg(target_email,msg):
  sender = 'xxxxx@163.com'  #发送信息的邮箱
  receivers = [target_email]  # 接收邮件，可设置为你的QQ邮箱或者其他邮箱
 
  # 三个参数：第一个为文本内容，第二个 plain 设置文本格式，第三个 utf-8 设置编码
  message = MIMEText(msg, 'plain', 'utf-8')
  subject = 'nvidia显卡监控'
  message['Subject'] = Header(subject, 'utf-8')
  #server = smtplib.SMTP('smtp.163.com', 587)
  server = smtplib.SMTP_SSL('smtp.163.com')#这部分需要去发送的邮箱账号去开启IMAP服务，获取登录授权码
  server.connect('smtp.163.com',465)
  #server.starttls()
  #server.ehlo()
  #xxxxxx是获取的登录授权码
  server.login(sender, "xxxxxx") 
  server.sendmail(sender, receivers, message.as_string()) 
  server.quit()

3.完整脚本nvidia.py

#-*-coding:GBK -*- 
import pynvml
pynvml.nvmlInit()
import time
import os
#from send_email import send_msg

import smtplib
from email.mime.text import MIMEText
from email.header import Header
 
def send_msg(target_email,msg):
  sender = 'xxxxx@163.com'
  receivers = [target_email]  # 接收邮件，可设置为你的QQ邮箱或者其他邮箱
 
  # 三个参数：第一个为文本内容，第二个 plain 设置文本格式，第三个 utf-8 设置编码
  message = MIMEText(msg, 'plain', 'utf-8')
  subject = 'nvidia显卡监控'
  message['Subject'] = Header(subject, 'utf-8')
  #server = smtplib.SMTP('smtp.163.com', 587)
  server = smtplib.SMTP_SSL('smtp.163.com')
  server.connect('smtp.163.com',465)
  #server.starttls()
  #server.ehlo()
  server.login(sender, "xxxxxxx") 
  server.sendmail(sender, receivers, message.as_string()) 
  server.quit()
  
  #smtpObj = smtplib.SMTP("localhost",1025)
  #try:   
  #    smtpObj.sendmail(sender, receivers, message.as_string())
  #    print("邮件发送成功")
  #except smtplib.SMTPException:
  #    print("Error: 无法发送邮件")


def watch_nvidia(nvidia_ids,min_memory):
  flag = [1 for i in nvidia_ids]
  for i in nvidia_ids:
    handle = pynvml.nvmlDeviceGetHandleByIndex(i)
    meminfo = pynvml.nvmlDeviceGetMemoryInfo(handle)
    print("card {} free memory is {}GB".format(i,meminfo.free * 1.0 /(1024**3)))
    if meminfo.free * 1.0 /(1024**3) > min_memory:
      flag[i-1]=0
    else:
      flag[i-1]=1
  if 0 in flag:
    free_num = 0
    for i in flag:
      if i == 0:
        free_num += 1
    return free_num
  else:
    print("no free card!")
    return -1

nvidia_ids = [0,1] # 显卡id
min_memory = 8 # 最小可用显存 GB
while True:
  flag = watch_nvidia(nvidia_ids,min_memory)
  if flag >= 1:
    send_msg("xxxxxxx@bjtu.edu.cn","{}张显卡空闲".format(flag))
    #os.system("sh veri.sh") # your command
    break
  time.sleep(10)

4.后台运行

nohup python nvidia.py >nvidia.out&