GIT version changelog:

2.31.0: 
      Two new ways to feed configuration variable-value pairs via
      environment variables have been introduced, and the way
      GIT_CONFIG_PARAMETERS encodes variable/value pairs has been tweaked
      to make it more robust.

Related GIT commits:

f9dbb64fadf599c588a39d2251bb3f9a2f7d572a  2021-01-12 13:27 +0100 Jeff King config: parse more robust format in GIT_CONFIG_PARAMETERS

So old

environment.h

#define CONFIG_DATA_ENVIRONMENT "GIT_CONFIG_PARAMETERS"
#define CONFIG_COUNT_ENVIRONMENT "GIT_CONFIG_COUNT"

List details about dangling commits (Git – Finding Lost/Dangling Commits:

git fsck --lost-found | grep "^dangling commit" | sed "s/^dangling commit //g" | xargs git show -s --oneline

Cleanup:

git reflog expire --expire=now --all
git gc --prune=now --aggressive
git repack -a -d

It is important to check if git configured with default settings. Especially forcing of creation of multiple smaller packs significantly increases repository size. I.e. changing one 1.2G pack into multiple 256MB packs let, increases size of packs to 4.3GB :(

git config --global -l
git config --system -l

Especially take attention for options like:

• deltaCacheSize = 10m
• packSizeLimit = 10m
• windowMemory = 10m

Command

git gc

and other cleaning commands requires whole repository to be read into memory. For big repos (like kernel.git) it requires to allocate over 8GB of RAM. Compression of objects is also performed in RAM, then packs are written onto disc. https://stackoverflow.com/questions/42175296/git-gc-uses-a-lot-of-memory-even-i-limited-it

To prevent touching existing packs and exhaust system memory paramter auto should be used.

git gc --auto

Another possibility is to mark some packs as keep by creating .keep files.

Remove unreachable objects:

git count-objects -v
git gc  --auto # this is enough to remove loose objects and repack repo
git count-objects -v
 
# more instruction
git repack -Ad # Remove dangling objects from packfiles
git prune # Remove dangling loose objects
git gc --aggressive --prune=now --force  || rm -f .git/objects/*/tmp_* && rm -f .git/objects/*/.tmp-*
 
git count-objects -v

git was killed by host monitor process or oom due to memory usage. Check syslog for OOM entries.

Solution:

[core]
    packedGitLimit = 32m
    packedGitWindowSize = 32m
    deltaCacheSize = 10m

Disable compression:

git config core.compression 0
git config core.loosecompression 0
git config pack.window 0

git command running under user 'git' with gitlab installed looks for configuration in following places:

• current git repo
• /etc/gitconfig
• /var/opt/gitlab/.config/git/config
• /var/opt/gitlab/.gitconfig

This is a user config gile, and file is generated form /opt/gitlab/embedded/cookbooks/gitlab/templates/default/gitconfig.erb

But if git binary from gitlab is used

sudo -u git -H "/opt/gitlab/embedded/bin/git" config -l

/opt/gitlab/embedded/bin/git --git-dir=/home/git-data/repositories/funny/linux.git config -l
/opt/gitlab/embedded/bin/git --git-dir=/home/git-data/repositories/funny/linux.git bundle create /tmp/linux.bundle --all

then, file /opt/gitlab/embedded/etc/gitconfig is used. To modify this file:

gitlab.rb

omnibus_gitconfig['system'] = {.
    "core"     => ["packedGitLimie = 32m", "packedGitWindowSze = 32m", "deltaCacheSize = 32m" ],
    "receive"  => ["fsckObjects = true"],
    "pack"     => ["deltaCacheSize = 32m", "packSizeLimit = 32m", "windowMemory = 32m", "threads = 2" ]
}

Test config:

gitlab-ctl show-config

Regenerate config:

gitlab-ctl reconfigure

sudo -u git bash
cd /var/opt/gitlab/git-data/repositories/{namespace}/{repository}.git
git gc --aggressive