Elastic Compute Service:Monitor Linux metrics with atop

Install atop

1. Log on to a Linux instance using Workbench.

2. Install atop:

   Alibaba Cloud Linux 4/3/2, CentOS 7/8, or Fedora

   # Update packages.
   sudo yum update -y
   
   # Install atop.
   sudo yum install atop -y

   Ubuntu or Debian

   # Update packages.
   sudo apt update -y
   
   # Install atop.
   sudo apt install atop -y

   CentOS Stream 9 or Rocky Linux 9

   # Update packages.
   sudo dnf update -y
   
   # Install the EPEL repository.
   sudo dnf install epel-release -y
   
   # Install atop.
   sudo dnf install atop -y

   openSUSE

   # Update packages.
   sudo zypper update -y
   
   # Install atop.
   sudo zypper install atop atop-daemon -y

3. Start the atop service:

   sudo systemctl start atop

4. Verify that the atop service started.

   Run sudo systemctl status atop. The active (running) status indicates success.

View real-time metrics

1. Run atop [sampling_interval_in_seconds] [number_of_samples] to enter interactive mode. The output has two sections: system resource overview and process details.

   # View with default settings (refreshes every 10 seconds).
   atop
   
   # View system metrics every 5 seconds.
   atop 5
   
   # Collect system metrics 30 times at 10-second intervals.
   atop 10 30
   
   # Collect metrics 10 times at 30-second intervals and write the results to a file.
   atop 30 10 > /tmp/atop.mem

   • System resource overview: Summarizes CPU, memory, swap, disk I/O, and network usage.

     

   • Process-level details: Shows resource consumption per process.

     

2. Use single-letter keys to switch views or sort processes in interactive mode.

   Key	Function	Purpose
   c	Display the full command line	View the startup parameters of a process.
   g	Generic view (default)	Summary of CPU, memory growth, and disk I/O.
   m/M	Memory view/Sort by memory	Analyze memory usage, page faults, and memory leaks.
   d/D	Disk view/Sort by disk	Identify processes with high disk read/write activity.
   n/N	Network view/Sort by network (requires you to install and load the netatop kernel module)	Identify processes with high network traffic.
   a	Aggregation view	Aggregate resource consumption of all threads or processes per program.
   C	Sort by CPU	Identify the most CPU-intensive processes.
   h	Show help	List all keyboard shortcuts.
   q	Quit	Exit interactive mode.

View historical metrics

The atop service periodically writes binary log files to the /var/log/atop/ directory, named in atop_YYYYMMDD format.

atop logs are binary files. Opening them with a text editor like cat, less, or vim displays unreadable characters.

Generate metric reports

atopsar extracts data from atop binary log files to generate system performance reports.

atopsar [options] [interval] [count]

Metric reference

Apply in production

• Disk space planning: atop log size depends on collection frequency, process count, and retention period. Check a single log file size with ls -lh /var/log/atop/ and multiply by LOGGENERATIONS to estimate total usage.

• Monitoring interval: For granular analysis, set LOGINTERVAL to 30 seconds. Shorter intervals increase disk I/O and CPU overhead of atop. Adjust based on your workload.

FAQ

1. Why do I get the error stat raw file: No such file or directory when running atop -r?

   This error means atop cannot find the log file for the specified date. Possible reasons:

   1. The atop service was not running on that date.

   2. The log was deleted by log rotation (LOGGENERATIONS exceeded).

   3. The queried date is in the future.

      Check /var/log/atop for available log files.

2. How can I reduce atop's disk space usage?

   Modify the configuration file and restart the atop service:

   1. Reduce the log retention period: Lower the value of LOGGENERATIONS in the configuration file, for example, from 28 to 14.

   2. Decrease the collection frequency: Increase the value of LOGINTERVAL in the configuration file, for example, from 600 to 1200.

3. Why can't I see per-process network traffic when I press the n key in atop's interactive mode?

   atop does not track per-process network traffic by default. Install and load the netatop kernel module to enable this.

   1. Install the kernel development package and compilation tools:

      sudo yum install -y kernel-devel dkms elfutils-libelf-devel

   2. Download the netatop source code:

      cd /usr/src/ && sudo wget https://www.atoptool.nl/download/netatop-3.2.2.tar.gz

   3. Extract the source code and enter the directory:

      sudo tar -zxvf netatop-3.2.2.tar.gz && cd netatop-3.2.2

   4. Build and install the module:

      sudo make && sudo make install

   5. Start the netatop service:

      sudo systemctl start netatop

   6. Verify that netatop is installed.

      Run atop and press n. A NET column in the process list confirms success.

      

4. Why can't I view the historical processes that caused the fault after restarting my ECS instance?

   By default, Linux does not record process-level historical resource usage, and restarting releases process information from memory. Without a monitoring tool installed beforehand, you cannot trace back to the fault scene before the restart, including issues such as system hangs, out-of-memory (OOM) errors, high kswapd0 load, and full I/O saturation.

   To be able to locate the root cause through historical logs when the next fault occurs, complete the following configuration in advance:

   Install atop as described in this topic and enable auto-start on boot (sudo systemctl enable atop) to ensure that atop continuously records process snapshots after the instance restarts. After configuration, when the next fault occurs, load historical logs with atop -r /var/log/atop/atop_YYYYMMDD to locate the root cause.