Reproducing ClusterFuzz bugs
If you've been assigned a ClusterFuzz bug but aren't sure what to do next, you're in the right place. If you want to help catch more bugs, you should also read up on developing new fuzzers and using ClusterFuzz.
Noticed a problem? File a bug using this template.
In the best case scenario, downloading the minimized test case and running it in chrome (or whatever binary you are testing) will be enough to reproduce the issue. Usually, you will also need to include the command line flags specified in the report as well.
When this doesn't work, you can also try downloading the exact build that the crash was found in via the "build" link in the ClusterFuzz report. If you are able to reproduce the issue using that, but not in your build, it may be a sign that you need to enable some kind of memory debugging tool, such as AddressSanitizer. The tool that you should will depend on the type of bug. See the common problems section for more information.
More complicated cases, such as test cases with gestures, may require you to use the ClusterFuzz local reproduction script. There are several quirks about that bot environment that this script is able to emulate which may also help you reproduce the crash consistently. Download the "local reproduction config" file from the report, and pass it as the --config argument to the script. If you haven't installed the script yet, see the instructions below. Unfortunately, it is Googler-only at the moment :(
If you still aren't able to reproduce the issue, you have a few options. For certain job types such as SyzyASan ones, a minidump may also be available. The report itself also includes some information about what may have gone wrong. If this is enough to work with, you could attempt a speculative fix. If there is nothing actionable in the report, simply mark it as WontFix.
- Test cases with gestures tend to be very difficult to reproduce. If you notice that a test case has gestures (look for the "interaction gestures" section at the top of the report), you should try using the local reproduction script.
- Some test cases do not reproduce correctly unless served over HTTP. Look for a "requires" section in the report, or a notification that the test case requires HTTP on the bug.
- You may not be able to reproduce all issues in a normal chromium build since ClusterFuzz makes heavy use of various memory debugging tools. Try using AddressSanitizer, SyzyASan, MemorySanitizer, LeakSanitizer, ThreadSanitizer, UndefinedBehaviorSanitizer, or whichever tool is being used in the report that has been assigned to you.
- Sometimes, a crash just isn't reproducible. If a crash is marked as unreproducible in ClusterFuzz and you're seeing similar results, don't spend too long re-running it. If the information in the report isn't enough to work with, don't be afraid to mark the bug as WontFix.
- Some tools, like the IPC fuzzer, have a more complicated setup process. In these cases, we try to provide additional help links in the report itself to documentation which may be useful.
If you weren't able to reproduce the issue using only the minimized test case and command line arguments from the report, you may want to try using the local reproduction script. It attempts to mimic the environment on the ClusterFuzz bots as closely as possible, and will attempt to run the test multiple times in case it's flaky.
See goto.google.com/clusterfuzz-repro for setup details.