Adding things to a git bare repository -

i know if have files in bare repository, can access them using git show head:path/to/file.

but can add new content bare repository without cloning , modifying working tree?

if add 1 file, 1 file in commit, aka added new , it's set in stone

there's several convenient ways add single-file commit tip of master branch in bare repo.

so appears need create blob object, attach tree, attach tree object commit.

all ways commit boil down doing that, it's question of how convenience commands suit purpose. git add creates blob , makes entry in index; git commit git write-tree adds new trees what's in index, , git commit-tree adds commit of top-level resulting tree, , git update-ref keep head date. bare repos have head commit, attached (aka symbolic ref for) branch master, . . .

so git's convenience commands doing want. 1 file, going easy.

say example files appear in ~server/data/logs/, bare repo you're using distribution @ ~server/repo.git, want committed files @ data/logs in repo, , want commit latest logfile:

#!/bin/sh cd ~server  # supply locations git ordinarily on own in working i.e. non-bare repos:  export git_dir=$pwd/repo.git                  # bare repos don't have defaults these export git_work_tree=$pwd                     # supply suit our purpose export git_index_file=$git_dir/scratch-index  # ...  # payload:  commit (only) latest file in data/logs:  git read-tree --empty                       # make index pretty, ,  git add data/logs/`ls -1t data/logs|sed q`  # everything's ordinary here - add ,  git commit -m'new logfile'                  # commit 

git read-tree loads index entries committed trees. it's underlies checkout , merge , reset , others i'm forgetting atm. here, want empty index start, hence --empty.

use push/pull/remote synchronize data while using tool available on every machine

you said "millions" of files on time, , if don't want full history distributed, rsync gather suspect might better bet. -- 1 @ time, 1 new file per minute, it'll take 2 years accumulate 1 million. so, ?

whatever, above procedure pretty efficiently extensible small-ish numbers of files per commit. bulk work there better ways.