Fixed deadlock issue related with MMDetWandbHook

[MilkClouds]

Co-authored-by: WangYudong yudong.wang@akane.waseda.jp

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand some items, don't worry, just make the pull request and seek help from maintainers.

Motivation

Please refer to #8145.
To summarize problem, when we use MMDetWandbHook after TextLoggerHook, problem such as deadlock or AssertionError: loss log variables are different across GPUs! occurs. These are because of same reason and I summarized the reason in #8145 (comment) here.

Since same problem occurs in mmseg open-mmlab/mmsegmentation#2137, I'm going to make almost-same PR on mmseg after merge of this PR.

Modification

Made MMDetWandbHook to clear runner.log_buffer on every process, including process for GPU 1,2,3...

BC-breaking (Optional)

Does the modification introduce changes that break the backward-compatibility of the downstream repos?
If so, please describe how it breaks the compatibility and how the downstream projects should modify their code to keep compatibility with this PR.

Use cases (Optional)

If this PR introduces a new feature, it is better to list some use cases here, and update the documentation.

Checklist

1. Pre-commit or other linting tools are used to fix the potential lint issues.
2. The modification is covered by complete unit tests. If not, please add more unit test to ensure the correctness.
3. If the modification has potential influence on downstream projects, this PR should be tested with downstream projects, like MMDet or MMCls.
4. The documentation has been modified accordingly, like docstring or example tutorials.

[MilkClouds]

There was already a similar commit on mmcls open-mmlab/mmpretrain#1242, which I didn't know! (this commit was only 5 days ago, too.)
I spent quite a long time trying to figure out the cause of the bug, but there was someone who already figured out the causes :(
Anyway, please merge this PR. Even though detailed implementation is different with open-mmlab/mmpretrain#1242, purpose is the same.

Also I guess we don't need to separate after_train_epoch, since runner's output buffer is cleared before epoch. I'll test once and update PR.

[codecov]

Codecov Report

Base: 64.15% // Head: 64.14% // Decreases project coverage by -0.01% ⚠️

Coverage data is based on head (df86a40) compared to base (31c8495).
Patch coverage: 66.66% of modified lines in pull request are covered.

❗ Current head df86a40 differs from pull request most recent head 36680cc. Consider uploading reports for the commit 36680cc to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##              dev    #9476      +/-   ##
==========================================
- Coverage   64.15%   64.14%   -0.02%     
==========================================
  Files         361      361              
  Lines       29583    29585       +2     
  Branches     5033     5033              
==========================================
- Hits        18980    18977       -3     
- Misses       9599     9601       +2     
- Partials     1004     1007       +3     

Flag	Coverage Δ
unittests	64.12% <66.66%> (-0.01%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
mmdet/core/hook/wandblogger_hook.py	15.20% <66.66%> (+0.28%)	⬆️
mmdet/utils/setup_env.py	88.00% <0.00%> (-4.00%)	⬇️
mmdet/models/roi_heads/mask_heads/maskiou_head.py	87.35% <0.00%> (-2.30%)	⬇️
mmdet/datasets/builder.py	52.08% <0.00%> (-1.05%)	⬇️

Help us with your feedback. Take ten seconds to tell us how you rate us. Have a feature suggestion? Share it here.

☔ View full report at Codecov.
📢 Do you have feedback about the report comment? Let us know in this issue.

[MilkClouds]

I've verified my commit, and merged two commit in one. plz check and merge this.