Best practices for masters and relays deployed on Linux

The most common cause for deteriorated performance (with frequent crash) of masters and relays on Linux is because the default values set for descriptors are very low. The following configuration checks help you improve the performance (and avoid crashes) of masters and relays on a Linux platform:

1. Connect to the device locally and launch the terminal
   OR
   Connect to the device using SSH.

2. Run the following commands as root (or as sudoer) to check the existing values:

   To check	Run command	Value check
   Maximum number of open files per user login	ulimit –Hn	If the returned value is less than 10240, run the commands in the step 3a.
   System-wide maximum number of open files	sysctl fs.file-max	If the value returned is less than 100000, Run the commands in the step 3b
   TCP max sync backlog	sysctl net.ipv4.tcp_max_syn_backlog	If the value returned is less than 5000, Run the commands in the step 3c.
   Read-ahead parameter	blockdev –getra /dev/sda	If the value returned is less than 4096, run the commands in the step 3d.

3. Depending on the returned values in step 2, run the following commands:

   Warning

   Exercise caution when running the following commands. If the commands are not run correctly, the performance of the device may deteriorate and further impact the stability of the device.

   1. To set the maximum number of open file descriptors per user login (value returned by the ulimit-HnRun command):

      # vi /etc/security/limits.conf
      
      root soft nofile 4096
      root hard nofile 10240

      Save the file (:q) and run the following command to validate the changes:

      # ulimit -Hn
      # ulimit -Sn

   2. To set the system-wide maximum number of open files:

      # vi /etc/sysctl.conf
      fs.file-max = 100000

      Save the file (:q) and run the following command to validate the changes:

      # sysctl -p

      Run the following command to double-check:

      # sysctl fs.file-max

   3. To set the TCP maximum sync backlog:

      # vi /etc/sysctl.conf
      net.ipv4.tcp_max_syn_backlog=5000

      Save the file (:q) and run the following command to validate the changes:

      # sysctl -p

      Run the following command to double-check:

      # sysctl net.ipv4.tcp_max_syn_backlog

   4. To set the  Read-Ahead  value to 4096 for each /dev/sd drive on your server, edit the /etc/rc.local file and add the following block at the end of the file (this is an example of a server with three drive entries - a, b, and c):

      blockdev --setra 4096 /dev/sda
      blockdev --setra 4096 /dev/sdb
      blockdev --setra 4096 /dev/sdc 
      

   5. Applying noatime attribute can also significantly improve the file I/O performance. To apply noatime attribute by editing the /etc/fstab file and replacing the 'defaults' with 'defaults,noatime', run the following command:

      # /etc/fstab
      /dev/mapper/centos-root /                       xfs     defaults,noatime 0 0
      UUID=a8f64424-1a43-4735-a20e-54a8f43304fe /boot xfs     defaults,noatime 0 0
      /dev/mapper/centos-home /home                   xfs     defaults,noatime 0 0
      /dev/mapper/centos-swap swap                    swap    defaults,noatime 0 0
4. Reboot the device.