Diagnosing Sudden Increase in MySQL File Handle Count

The article presents a step‑by‑step experiment to reproduce a rapid increase in MySQL file handles by creating and enlarging tables a and b , adding columns, and executing a query that triggers a hash join.

During the query execution the author monitors MySQL’s file‑handle count and observes a continuous rise, indicating that some internal operation is opening many descriptors.

Linux perf is then employed to trace the open system calls. The perf script output shows that the handles are associated with the hash‑join phase, where probe rows are written to temporary files on disk.

The article references the MySQL worklog describing the hash‑join implementation, confirming that temporary spill files can cause additional file descriptors.

To further verify the source of the handles, the author queries the Performance Schema with select * from performance_schema.file_instances; , which lists the active file instances and confirms their relation to the hash‑join operation.

The conclusion is that the rising file‑handle count is caused by the hash‑join’s need to spill data to disk, and deeper analysis with perf is required to isolate the exact step.