dbslower - Trace MySQL/PostgreSQL server queries slower than a threshold. dbslower [-v] [-p PID [PID ...]] [-m THRESHOLD] {mysql,postgres} This traces queries served by a MySQL or PostgreSQL server, and prints those that exceed a latency (query time) threshold. By default a threshold of 1 ms is used. This uses User Statically-Defined Tracing (USDT) probes, a feature added to MySQL and PostgreSQL for DTrace support, but which may not be enabled on a given installation. See requirements. Since this uses BPF, only the root user can use this tool. CONFIG_BPF, bcc, and MySQL server with USDT probe support (when configuring the build: -DENABLE_DTRACE=1) or PostgreSQL server with USDT probe support (when configuring the build: --enable-dtrace). -h Print usage message.

-p PID Trace this PID. If no PID is specified, the tool will attempt to automatically detect the MySQL or PostgreSQL processes running on the system.

-m THRESHOLD Minimum query latency (duration) to trace, in milliseconds. Default is 1 ms.

{mysql,postgres} The database engine to trace.

Trace MySQL server queries slower than 1 ms: # dbslower mysql

Trace slower than 10 ms for PostgreSQL in process 408: # dbslower postgres -p 408 -m 10

TIME(s) Time of query start, in seconds.

PID Process ID of the traced server.

MS Milliseconds for the query, from start to end.

QUERY Query string, truncated to 256 characters.

This adds low-overhead instrumentation to queries, and only emits output data from kernel to user-level if they query exceeds the threshold. If the server query rate is less than 1,000/sec, the overhead is expected to be negligible. If the query rate is higher, test to gauge overhead. This is from bcc. https://github.com/iovisor/bcc

Also look in the bcc distribution for a companion _examples.txt file containing example usage, output, and commentary for this tool.

Linux Unstable - in development. Sasha Goldshtein, Brendan Gregg biosnoop(8), mysqld_qslower(8), dbstat(8)