Using Percona Toolkit’s pt‑stalk for MySQL Monitoring and Diagnostics

The author shares a practical guide to the Percona Toolkit tool pt‑stalk , which collects OS and MySQL diagnostic information when specified trigger conditions are met.

Overview : After a previous article on MySQL row‑lock timeout monitoring, the author discovered that pt‑stalk can monitor those locks and many other metrics.

Introduction : pt‑stalk is part of the Percona Toolkit suite, alongside familiar tools such as pt‑query‑digest and pt‑online‑schema‑change . It is a shell script that gathers CPU, I/O, memory, disk, network data at the OS level and row‑lock wait, session, replication, and status variables at the MySQL level.

Key Features : OS metrics: CPU, I/O, memory, disk, network. MySQL metrics: row‑lock wait, session connections, replication status, various status variables.

Usage : The tool runs as a background service monitoring MySQL and triggers data collection based on configurable parameters:

function – default status (monitors SHOW GLOBAL STATUS ) or processlist (monitors SHOW PROCESSLIST ).

variable – default Threads_running ; can be set to any status variable such as Threads_connected .

threshold – default 25; the metric must exceed this value to trigger.

cycles – default 5; the condition must be observed consecutively this many times.

Connection parameters: host , port , user , password , socket .

Other options: iterations , run‑time , sleep , interval , dest , retention‑time , daemonize , log , collect (including collect‑gdb , collect‑strace , collect‑tcpdump ).

Example command :

pt‑stalk --function status --variable Threads_connected --threshold 2500 --daemonize --user=root --password=######

Scenarios : Monitor Threads_connected and trigger when it exceeds 1000 for five consecutive checks. Monitor Threads_running with a threshold of 500. Run a one‑off collection for 60 seconds without any trigger, useful for quick problem capture.

Monitoring Output : pt‑stalk creates many files named after the commands that generated them, including lock‑waits and transactions which are valuable for diagnosing row‑lock issues.

pt‑sift : The companion tool pt‑sift can summarize the data collected by pt‑stalk , providing a concise report of the most relevant diagnostics.

Conclusion : pt‑stalk is a lightweight, trigger‑based monitoring solution ideal for environments without a full‑blown monitoring system, especially for temporary MySQL troubleshooting. Its main limitations are that it works best on local MySQL instances and supports only a single trigger condition at a time.