MIRI check is failing on master

[alamb]

Describe the bug
The MIRI check has started to fail on master:

https://github.com/apache/arrow-rs/commits/master

Example failure: https://github.com/apache/arrow-rs/runs/4023917285?check_suite_focus=true

To Reproduce
Run CI MIRI test on master

Expected behavior
Test should pass

Additional context
The MIRI version was recently updated in #875 by @jimexist

[jimexist]

@alamb i merely bumped the version.

i suspect it's because the 1.56 ver won't play nice with miri

i tried to fix it in https://github.com/apache/arrow-rs/pull/591/files#diff-621a5063e07f71e640bc5af8989b652fb7f6f183e7ef81b819a64f9670bea7f0R2044

[alamb]

I could not reproduce this locally (on MacOS):

Here is what i tried:

rustup toolchain install nightly-2021-10-23
rustup component add miri --toolchain nightly-2021-10-23
MIRIFLAGS="-Zmiri-disable-isolation" cargo +nightly-2021-10-23  miri test -p arrow -- --skip csv --skip ipc --skip json

Results in

...
test src/record_batch.rs - record_batch::RecordBatch::try_from_iter (line 284) ... ok
test src/record_batch.rs - record_batch::RecordBatch::try_from_iter_with_nullable (line 321) ... ok
test src/compute/kernels/window.rs - compute::kernels::window::shift (line 32) ... ok
test src/util/test_util.rs - util::test_util::arrow_test_data (line 73) ... ok
test src/util/test_util.rs - util::test_util::parquet_test_data (line 95) ... ok
test src/util/string_writer.rs - util::string_writer (line 25) ... ok
test src/compute/kernels/comparison.rs - compute::kernels::comparison::like_utf8 (line 239) ... ok

test result: ok. 117 passed; 0 failed; 0 ignored; 0 measured; 14 filtered out; finished in 5.76s


Compilation finished at Sat Oct 30 09:58:38

Going to try it on Linux next

[jimexist]

Nor can I on MacBook Pro intel CPU

[alamb]

🤔 the same commands passed for me on Ubuntu 20.04 on x86 as well

test src/util/string_writer.rs - util::string_writer (line 25) ... ok
test src/compute/kernels/comparison.rs - compute::kernels::comparison::like_utf8 (line 239) ... ok

test result: ok. 117 passed; 0 failed; 0 ignored; 0 measured; 14 filtered out; finished in 5.85s

[jimexist]

Can it because of memory consumption?

[alamb]

Can it because of memory consumption?

"Great minds think alike" -- lol I tried setting to --test-threads=1 in this commit to test that theory but that didn't seem to work

0dd0d62

However, I have had better luck clearing out the old cache. I'll get a new PR up shortly

[alamb]

It appears to have resurfaced: https://github.com/apache/arrow-rs/commits/master

I am going to try to disable caching entirely for MIRI job and see if that looks promising

[alamb]

Still happening: https://github.com/apache/arrow-rs/runs/4069047142?check_suite_focus=true

Passes sometimes on rerun....

[alamb]

I did an experiment and confirmed the symptoms we are seeing are the same symptoms as happens when MIRI is killed on github runner due to OOM: https://github.com/alamb/oom_canary#question-what-is-the-symptom-when-cargo-miri-exceeds-its-memory-limits-on-github-runner