fuzzbench icon indicating copy to clipboard operation
fuzzbench copied to clipboard
verified publisher icon google

Measurers use an unreasonable amount of memory

Open addisoncrump opened this issue 11 months ago • 2 comments

See issue title. When I was re-running the measurer for an experiment for which the measurers terminated early, and did not specify the runners, I OOM'd a 750GB RAM machine in a matter of minutes. The measurers stalled and ultimately returned to a stable state.

Screenshot of memory usage. Python processes are using multiple hundreds of GB of RAM each.

addisoncrump avatar Feb 03 '25 12:02 addisoncrump

Actually, I just had this reproduce with the number of runners specified.

addisoncrump avatar Feb 03 '25 12:02 addisoncrump

This seems to be related to the systemd fuzzer -- certain fuzzers produce inputs which cause it to OOM, but the OOM killer doesn't seem to always activate while the measurement is running.

addisoncrump avatar Feb 03 '25 13:02 addisoncrump