About Clustefuzz corpus assign strategy

[gtt1995]

In the clusterfuzz corpus strategy, can all seeds be arranged in libfuzzer_corpus according to size, and each bot only takes one copy from it for execution?

The initial strategy is to take out all the seeds each time. I think this large number of seeds will cause pressure within a single bot, and the use of the same initial corpus by each bot with the same engine will lead to a certain degree of repetitive work.

Here's a job to simply evaluate the advantages of this strategy.apply to libFuzzer

Since I don't know how to evaluate the performance of this improvement, can you guide me where the code changes are accurate? @jonathanmetzman @inferno-chromium @oliverchang @mbarbella-chromium Huge thanks for your helps!

[gtt1995]

It automatically just works, just define the libfuzzer, honggfuzz, afl job types containing fuzz binaries with same names and then they just share the same corpora automatically, see "CORPUS_FUZZER_NAME_OVERRIDE = libFuzzer" in afl and honggfuzz templates here - https://github.com/google/clusterfuzz/blob/master/src/local/butler/scripts/setup.py#L57

Is it possible to sort seeds in this shared corpus to find more effective seeds? This has been proven to be very effective in libFuzzer-FORK. Although this is shared under different engines.

[gtt1995]

My simple method is: first sort the seeds in the corpus according to the size of the seeds, and then divide them into multiple groups evenly. Each bot takes out a copy as its own initial corpus without repetition, and loops in turn. Can anyone give me some advice? Thanks!

[gtt1995]

@jonathanmetzman @oliverchang, Sorry to bother you, looking forward to your reply and email.