memory usage issue.

[hoperok5]

Question

MediaMTX memory usage is increasing.

About 300 source streams are registered in the configuration, and they support RTSP (tcp only) and HLS.
Of the 300 streams, only 70 are being used as live-pull streams, and multiple clients are accessing them frequently.
Testing was conducted for approximately 20 hours, and the pprof data file is attached.
We need to determine whether this is an internal issue with MediaMTX or a GC optimization issue.

Test Environment

• MediaMTX-1.14.0
• Docker Container: Ubuntu-24.04
• Host: CentOS 7.6

Memory Usage

pprof output file
pprof.mediamtx-250911.alloc_objects.alloc_space.inuse_objects.inuse_space.001.pb.gz
pprof.mediamtx-250911.goroutine.001.pb.gz
pprof.mediamtx-250911.samples.cpu.001.pb.gz

goroutine

Heap

Profile

[aler9]

Thank you very much for providing pprof reports, from what you provided it's evident that there's a memory leak, in particular counterdumper.Counterdumper is not released correctly and there are over 500.000 instances of it. I'll see what i can do.

[aler9]

this is fixed by #4967

[hoperok5]

@aler9
Thank you for your efforts.

I checked again with version 1.15.0.
However, the server is still showing symptoms that suggest a memory leak, although the memory increase is smaller than with this version.

[aler9]

@hoperok5 system memory consumption is not an indicator of a memory leak since Go is a garbage collector-based programming language, and memory is released only when necessary, and that happens when some percentage is reached (i don't remember if it's 50% or 75%). Only pprof reports are valid tools for debugging. If you have further issues, open a bug report and provide new pprof reports.

[hoperok5]

Thank you for your reply.

For reference, the memory usage information I posted is container memory usage.
I expected to be able to verify this using the previously posted pprof data.
I will collect and provide the pprof data once I conduct further testing.

Also, I understand that the Go garbage collector operates periodically based on the process's memory usage. I don't understand the statement that it operates when 50% or more of the total memory is consumed.