syzkaller
syzkaller copied to clipboard
google
dashboard/app: capture and report probability of reproducing a bug
Some repros trigger the bug immediately, some after some time, or not in each run. Several users asked for some kind of "how hard it is to re-trigger the crash using this repro" thing. E.g. what's average runtime? Multi-threaded/process repros are also some kind of signal.
Also, how much percents of Multi-threaded bugs can reproduce by syzbot by now?
@daydayup40 I don't have such info. To get this number, one needs to go over all syzbot reported bugs and mark them as race/norace (https://syzkaller.appspot.com, https://syzkaller.appspot.com?fixed=upstream).
Besides threaded/non-threaded, it can also make sense to run it multiple times and observe how many times we reproduced it. We could also capture set of all crashes that the repro produces. But then we need to figure out where/how to present this. This info may also be useful during bisection. We currently use 10 VMs. We could use less for reliable bugs and more for unreliable bugs.
