Add fuzz inputs from my servers and oss-fuzz

maflcko commented at 4:46 pm on October 17, 2023: contributor

Doing this in a single commit to allow the fuzz engine to create a smaller set of files.

murchandamus approved

murchandamus commented at 5:50 pm on October 17, 2023: contributor

LGTM, appears to just add 30k odd new fuzz inputs

ACK 0dbdd5523d4c67fc185f174aa7dfbcc81e10ad6b

maflcko commented at 5:57 pm on October 17, 2023: contributor

You are adding around 20k, so I wonder if there is an overlap. Should I redo with your folder added?

murchandamus commented at 6:46 pm on October 17, 2023: contributor

Sure, feel free to pull it in, that’s what I pushed the branches for! :)

maflcko commented at 9:40 am on October 18, 2023: contributor

Ok, running that now.

Also, it looks like we are nearing the time limit here.

So, I’d say to merge this and then right after open the pull request to delete-nonreduced.sh?

maflcko commented at 9:41 am on October 18, 2023: contributor

Re-running CI: https://cirrus-ci.com/task/5867579709849600

murchandamus commented at 12:19 pm on October 18, 2023: contributor

Ok, running that now.

Also, it looks like we are nearing the time limit here.

So, I’d say to merge this and then right after open the pull request to delete-nonreduced.sh?

Approach ACK

maflcko commented at 3:45 pm on October 18, 2023: contributor

Hmm, for some reason I am running out of disk space on my tmpfs /tmp. I’ll run again over night and push tomorrow morning, if it works :smiling_face_with_tear:

maflcko commented at 7:22 am on October 19, 2023: contributor

Now the task is DONE, but it failed (:smiling_face_with_tear:)

0[#131072](/bitcoin-core-qa-assets/131072/)	pulse  cov: 3399 exec/s: 672 rss: 219Mb
1[#262144](/bitcoin-core-qa-assets/262144/)	pulse  cov: 6464 exec/s: 579 rss: 219Mb
2[#451694](/bitcoin-core-qa-assets/451694/)	DONE   cov: 7097 exec/s: 422 rss: 316Mb
3MERGE-OUTER: successful in 1 attempt(s)
4MERGE-OUTER: the control file has 4052306816 bytes
5==1059184== ERROR: libFuzzer: out-of-memory (used: 2084Mb; limit: 2048Mb)
6   To change the out-of-memory limit use -rss_limit_mb=<N>
7
8SUMMARY: libFuzzer: out-of-memory

Add fuzz inputs from my servers, murch, and oss-fuzz 00c956e4a9

maflcko commented at 9:46 am on October 19, 2023: contributor

Ok, now it passed, with this patch applied to avoid out-of-memory and out-of-storage:

 0diff --git a/test/fuzz/test_runner.py b/test/fuzz/test_runner.py
 1index 9d64591111..591689b11c 100755
 2--- a/test/fuzz/test_runner.py
 3+++ b/test/fuzz/test_runner.py
 4@@ -278,6 +278,8 @@ def merge_inputs(*, fuzz_pool, corpus, test_list, src_dir, build_dir, merge_dirs
 5         args = [
 6             os.path.join(build_dir, 'src', 'test', 'fuzz', 'fuzz'),
 7             '-set_cover_merge=1',
 8+            '-rss_limit_mb=8000',
 9+            f'-merge_control_file=/run/media/marco/scratch_fde/fuzz/A_m_{t}.txt',
10             # set_cover_merge is used instead of -merge=1 to reduce the overall
11             # size of the qa-assets git repository a bit, but more importantly,
12             # to cut the runtime to iterate over all fuzz inputs [0].

Revert i2p fuzz inputs to work around crash for now c764d6c784

maflcko requested review from murchandamus on Oct 19, 2023

murchandamus commented at 1:30 pm on October 19, 2023: contributor

I verified that the second commit only modifies a bunch of I2P. Both the first and the second commit’s I2P seeds pass on my machine:

0➜  qa-assets git:(main) ✗ git reset --hard maflcko/main
1Updating files: 100% (36457/36457), done.
2HEAD is now at c764d6c7842 Revert i2p fuzz inputs to work around crash for now
3➜  build git:(2023-10-refactor-coinselector-tests) ✗ FUZZ=i2p src/test/fuzz/fuzz ../../qa-assets/fuzz_seed_corpus/i2p                                                                     <<<
4i2p: succeeded against 1247 files in 2s.
5➜  qa-assets git:(main) ✗ git reset --hard 00c956e4a9c555da5640946ec28cdd5e501c0554
6HEAD is now at 00c956e4a9c Add fuzz inputs from my servers, murch, and oss-fuzz
7➜  build git:(2023-10-refactor-coinselector-tests) ✗ FUZZ=i2p src/test/fuzz/fuzz ../../qa-assets/fuzz_seed_corpus/i2p
8i2p: succeeded against 1256 files in 2s.

:shrug:

The first commits seems to add a bunch of files into the fuzz_seed_corpus directly instead of one of the subfolders. Any idea what may have happened there?

maflcko commented at 1:35 pm on October 19, 2023: contributor

The first commits seems to add a bunch of files into the fuzz_seed_corpus directly instead of one of the subfolders. Any idea what may have happened there?

Can you give an example?

maflcko commented at 3:32 pm on October 19, 2023: contributor

Both the first and the second commit’s I2P seeds pass on my machine:

That also seems funny. You could reproduce yesterday (at least in your folder), I can reproduce locally, and both CI tasks failed on your pull requests with this error.

murchandamus commented at 5:24 pm on October 19, 2023: contributor

Sorry, PEBCAK on the files. In my tig window, it dropped the fuzz target folder on some files in the change list, but not the others. I didn’t realize that it would present me different levels of the path in one list. :facepalm: .

 0➜  qa-assets git:(main) ✗ git reset --hard maflcko/main                                   
 1HEAD is now at c764d6c7842 Revert i2p fuzz inputs to work around crash for now
 2➜  bitcoin git:(master) ✗ cd ../bitcoin/build                                            
 3➜  build git:(master) ✗ FUZZ=i2p src/test/fuzz/fuzz ../../qa-assets/fuzz_seed_corpus/i2p
 4i2p: succeeded against 1247 files in 27s.
 5➜  build git:(master) ✗ cd ../../qa-assets
 6➜  qa-assets git:(main) ✗ git reset --hard 00c956e4a9c555da5640946ec28cdd5e501c0554       
 7HEAD is now at 00c956e4a9c Add fuzz inputs from my servers, murch, and oss-fuzz
 8➜  qa-assets git:(main) ✗ cd -                                                            
 9~/Workspace/bitcoin/build
10➜  build git:(master) ✗ FUZZ=i2p src/test/fuzz/fuzz ../../qa-assets/fuzz_seed_corpus/i2p
11i2p: succeeded against 1256 files in 26s.
12➜  build git:(master) ✗

I don’t know why it doesn’t crash on my machine but I can not replicate it crashing with either of your two commits here.

maflcko commented at 7:55 pm on October 19, 2023: contributor

In any case, let’s move discussion to https://github.com/bitcoin/bitcoin/issues/28665

maflcko commented at 8:18 pm on October 19, 2023: contributor

Which fuzzing engine are you using? Can you try with libFuzzer?

I’ve never seen this command and output:

0FUZZ=i2p src/test/fuzz/fuzz ../../qa-assets/fuzz_seed_corpus/i2p
1i2p: succeeded against 1247 files in 27s.

maflcko commented at 7:48 am on October 20, 2023: contributor

For reference, this increases the coverage from 63.6% (https://maflcko.github.io/b-c-cov/fuzz.coverage/index.html , unstable link) to 64.4% (https://drahtbot.space/host_reports/DrahtBot/reports/coverage_fuzz/monotree/091d29c49590d31d/c764d6c7842481a2/fuzz.coverage/index.html)

fanquake merged this on Oct 20, 2023

fanquake closed this on Oct 20, 2023

hebasto commented at 3:34 pm on October 21, 2023: member

Could this PR cause CI timeouts?

See: https://github.com/bitcoin/bitcoin/issues/28703.

fanquake referenced this in commit 0f15db0ec9 on Oct 23, 2023

Frank-GER referenced this in commit 97900aaaf7 on Nov 28, 2023

Add fuzz inputs from my servers and oss-fuzz #155