BCC – Dynamic Tracing Tools for Linux Performance Monitoring

BPF CompilerCollection (BCC)这个工具集包含很多用来观测内核性能的工具，全部使用eBPF，并且提供了python的外部编程能力。本文前序文章： 张亦鸣: eBPF 简史

本文转载自：

https://www.tecmint.com/bcc-best-linux-performance-monitoring-tools/

关注Linuxer，订阅源源不断的一线技术文章

BCC (BPF Compiler Collection) is a powerful set of appropriate tools and example files for creating resourceful kernel tracing and manipulation programs. It utilizes extended BPF (Berkeley Packet Filters), initially known as eBPF which was one of the new features in Linux 3.15.

BCC/BPF – Dynamic Tracing Tools for Linux Performance Monitoring

Practically, most of the components used by BCC require Linux 4.1 or above, and its noteworthy features include:

1. Requires no 3rd party kernel module, since all the tools work based on BPF which is built into the kernel and BCC uses features added in Linux 4.x series.

2. Enables observation of software execution.

3. Comprises of several performance analysis tools with example files and man pages.

Best suited for advanced Linux users, BCC makes it easy to write BPF programs using kernel instrumentation in C, and front-ends in Python and lua. Additionally, it supports multiple tasks such as performance analysis, monitoring, network traffic control plus lots more.

How To Install BCC in Linux Systems

Remember that BCC uses features added in Linux kernel version 4.1 or above, and as a requirement, the kernel should have been compiled with the flags set below:

To check your kernel flags, view the file /proc/config.gz or run the commands as in the examples below:

After verifying kernel flags, it’s time to install BCC tools in Linux systems.

On Ubuntu 16.04

Only the nightly packages are created for Ubuntu 16.04, but the installation instructions are very straightforward. No need of kernel upgrade or compile it from source.

$ echo "deb [trusted=yes] https://repo.iovisor.org/apt/xenial xenial-nightly main" | sudo tee /etc/apt/sources.list.d/iovisor.list

$ sudo apt-get update

$ sudo apt-get install bcc-tools

On Ubuntu 14.04

Begin by installing a 4.3+ Linux kernel, from http://kernel.ubuntu.com/~kernel-ppa/mainline.

As an example, write a small shell script “bcc-install.sh” with the content below.

Note: update PREFIX value to the latest date, and also browse the files in the PREFIX url provided to get the actual REL value, substitute them in the shell script.

Save the file and exit. Make it executable, then run it as shown:

$ chmod +x bcc-install.sh$ sh bcc-install.sh

Afterwards, reboot your system.

$ reboot

Next, run the commands below to install signed BCC packages:

$ sudo apt-key adv --keyserver keyserver.ubuntu.com --recv-keys D4284CDD

$ echo "deb https://repo.iovisor.org/apt trusty main" | sudo tee /etc/apt/sources.list.d/iovisor.list

$ sudo apt-get update

$ sudo apt-get install binutils bcc bcc-tools libbcc-examples python-bcc

How To Use BCC Tools in Linux Systems

All the BCC tools are installed under /usr/share/bcc/tools directory. However, you can alternatively run them from the BCC Github repository under /tools where they end with a .py extension.

We shall cover a few examples under – monitoring general Linux system performance and networking.

Trace open() syscalls

Let’s start by tracing all open() syscalls using opensnoop. This enable us tell us how various applications work by identifying their data files, config files and many more:

Summarize Block Device I/O Latency

In this example, it shows a summarized distribution of disk I/O latency using biolatecncy. After executing the command, wait for a few minutes and hit Ctrl-C to end it and view the output.

Trace New Processes via exec() Syscalls

In this section, we shall move to tracing new processes in execution using execsnoop tool. Each time a process is forked by fork() and exec() syscalls, it is shown in the output. However, not all processes are captured.

Trace Slow ext4 Operations

Using ext4slower to trace the ext4 file system common operations that are slower than 10ms, to help us identify independently slow disk I/O via the file system.

It only outputs those operations that exceed a threshold:

Trace Block Device I/O with PID and Latency

Next off, let’s dive into printing a line per disk I/O each second, with details such as process ID, sector, bytes, latency among others using biosnoop:

Trace Page Cache hit/miss Ratio

Thereafter, we proceed to using cachestat to displays one line of summarized statistics from the system cache every second. This enables for system tuning operations by pointing out low cache hit ratio and high rate of misses:

Trace TCP Active Connections

Monitoring TCP connections every second using tcpconnect. Its output includes source and destination address, and port number. This tool is useful for tracing unexpected TCP connections, thereby helping us to identify inefficiencies in application configurations or an attacker.

All the tools above can also be used with various options, to enable the help page for a given tool, make use of the -h option, for example:

Trace Failed exec()s Syscalls

To trace failed exec()s syscalls, employ the -x option with opensnoop as below:

Trace Particular Process Functions

The last example below demonstrates how to execute a custom trace operation. We are tracing a particular process using its PID.

First determine the process ID:

$ pidof firefox

15437

Later on, run the custom trace command. In the command below: -p specifies the process ID, do_sys_open() is a kernel function that is traced dynamically including its second argument as a string.

Summary

BCC is a powerful and easy-to-use toolkit for various System administration tasks such as tracing system performance monitoring, tracing block device I/O, TCP functions, file system operations, syscalls, Node.js probes, plus lots more. Importantly, it ships in with several example files and man pages for the tools to guide you, making it user friendly and reliable.

Last but not least, you can get back to us by sharing your thoughts about the subject, ask questions, make useful suggestions or any constructive feedback via the comment section below.

For more information and usage visit: https://iovisor.github.io/bcc/

更多精彩内容，尽在阅读原文