Clean committed big file in Git

If you commit a big file in git, whose file size is larger than the pre-defined github setting (Not standard github with LFS, maybe your own enterprise github), how could you follow-up?

Two common ways for this:

1. git reset

Suitable case: the big files are set only in the latest commit, in that commit, there is nothing else, only with incorrectly committed big file.

git log # to find the nearest logging
git reset --hard <6bits commit log ID>

2. git filter-branch

Suitable case: the big files are set only in the latest commit, but in that commit, besides these big files, there are also useful information. It's to say that you can restore to one single hard-reset operation for removal of big files.

Find all of relevant files for that commit. take my case as example, if you commit the following files in commit history for master, which blocked your commit to remote github.

• MobileUser/FeatureRefine/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz
• FeatureRefine/object_detection/faster_rcnn_resnet101_coco_2018_01_28/frozen_inference_graph.p
• MobileUser/FeatureRefine/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz
• MobileUser/FeatureRefine/object_detection/ssd_mobilenet_v1_coco_2017_11_17/frozen_inference_graph.pb
• MobileUser/FeatureRefine/object_detection/test_ckpt/ssd_inception_v2.pb

We can use 'git filter-branch --tree-filter ' to remove file from commit history, compressed git log and recommit again. As commit history setting, you may need to do git push --force, when doing the final commit.

Script as follow:

git filter-branch --tree-filter 'rm -rf MobileUser/FeatureRefine/object_detection/faster_rcnn_resnet101_coco_2018_01_28.tar.gz' HEAD
git filter-branch --tree-filter 'rm -rf MobileUser/FeatureRefine/object_detection/faster_rcnn_resnet101_coco_2018_01_28/frozen_inference_graph.p' -f HEAD
git filter-branch --tree-filter 'rm -rf MobileUser/FeatureRefine/object_detection/ssd_mobilenet_v1_coco_2017_11_17.tar.gz' -f HEAD
git filter-branch --tree-filter 'rm -rf MobileUser/FeatureRefine/object_detection/ssd_mobilenet_v1_coco_2017_11_17/frozen_inference_graph.pb' -f HEAD
git filter-branch --tree-filter 'rm -rf MobileUser/FeatureRefine/object_detection/test_ckpt/ssd_inception_v2.pb' -f HEAD
git reflog expire --all &&  git gc --aggressive --prune
git push origin master --force