fuzzbench icon indicating copy to clipboard operation
fuzzbench copied to clipboard

Published 20 hours ago • verified publisher icon google

Minimize Corpus Seed

Open ardier opened this issue 11 months ago • 6 comments

Re-opening this experiment after minor correction. This experiment builds on our previous using the freetype2 benchmark with a custom seed corpus.

Please run the following command to run the experiment: /gcbrun run_experiment.py -a --experiment-config /opt/fuzzbench/service/experiment-config.yaml --experiment-name 2025-01-01-freetype2_ftfuzzer-mutants --fuzzers afl aflplusplus libfuzzer --benchmarks freetype2_ftfuzzer

@DonggeLiu @jonathanmetzman Could you please have a look?

Thank you kindly in advance.

ardier avatar Jan 04 '25 20:01 ardier

Sorry about the delay, @ardier. I just returned to work.

DonggeLiu avatar Jan 22 '25 05:01 DonggeLiu

/gcbrun run_experiment.py -a --experiment-config /opt/fuzzbench/service/experiment-config.yaml --experiment-name 2025-01-22-freetype2_ftfuzzer-mutants --fuzzers afl aflplusplus libfuzzer --benchmarks freetype2_ftfuzzer

DonggeLiu avatar Jan 22 '25 05:01 DonggeLiu

Hi @DonggeLiu.

Is there a way to access the logs? Compared to 2060, I see a few major differences:

  1. The seeds are in a .tar.gz here.
  2. The tarball might be larger.
  3. I am keeping TRT fonts here.

But, apart from that, it looks like a similar experiment.

I'd appreciate any insight you could provide on this.

ardier avatar Jan 28 '25 12:01 ardier

Wait I just noticed that the experiment did not run because the name is invalid:

experiment.run_experiment.ValidationError: Experiment name "2025-01-22-freetype2_ftfuzzer-mutants" is invalid. Must match: "^[a-z0-9-]{0,30}$"
  1. It has 37 chars, but should be within 30.
  2. It should not contain _.

A better name could be 2025-01-22-freetype2-mts. I will re-run it below.

Is there a way to access the logs?

Fuzzer logs can be accessed from the experiment data dir (https://storage.googleapis.com/fuzzbench-data/index.html?prefix=/)

Cloud logs cannot be publicly accessed unfortunately, a possible workaround is writing FuzzBench logs to your fuzzer log (I have not tested this though).

DonggeLiu avatar Jan 28 '25 23:01 DonggeLiu

/gcbrun run_experiment.py -a --experiment-config /opt/fuzzbench/service/experiment-config.yaml --experiment-name 2025-01-22-freetype2-mts --fuzzers afl aflplusplus libfuzzer --benchmarks freetype2_ftfuzzer

DonggeLiu avatar Jan 28 '25 23:01 DonggeLiu

Hello. I am not able to see the output of the experiment (presumably it failed again). Could you please check again?

ardier avatar Feb 03 '25 16:02 ardier