git post-receive template should use archive, not clone

[jyaworski]

Currently, the post-receive template uses git clone to populate environments. Using git archive would greatly reduce the space used, as it's close to svn export. Using my checkout of this repo:

joseph.yaworski@jyaworski-mbpr ~/github/puppet-puppet (git)-[master] % du -hs .
8.8M    .
joseph.yaworski@jyaworski-mbpr ~/github/puppet-puppet (git)-[master] % git archive --format tar --prefix export/ HEAD | tar -x
joseph.yaworski@jyaworski-mbpr ~/github/puppet-puppet (git)-[master] % du -sh export 
464K    export

Are there concerns using archive rather than clone?

Edit:
To check out a specific branch to a specific dir:

git archive --format tar --prefix branchname/ branchname: | tar -x -C /etc/puppet/environments

[ekohl]

I believe git clone on a local filesystem uses hardlinks so I wonder how much it actually uses.

[jyaworski]

Earlier, cloning from a local filesystem using the existing template, the first du (including the .git) was around 151M or so, with the .git being something like 80M of that.

[jyaworski]

I have actual stats now:

This is doing a traditional git clone from an upstream not on the machine.

[root@puppet1 ~]# du -sh /etc/puppet/environments/production/
148M    /etc/puppet/environments/production/
[root@puppet1 ~]# du -sh /etc/puppet/environments/production/.git
85M     /etc/puppet/environments/production/.git

This is cloning from the local /opt/puppet.git.

[root@puppet1 ~]# du -sh /etc/puppet/environments/testpostgres/
404M    /etc/puppet/environments/testpostgres/
[root@puppet1 ~]# du -sh /etc/puppet/environments/testpostgres/.git
340M    /etc/puppet/environments/testpostgres/.git

[ekohl]

This didn't get that much attention. I'm wondering if we should look at supporting r10k out of the box, but there are no packages for that on most distros. Thoughts?

[jyaworski]

@ekohl we could. Add a dependency on puppet-r10k. https://forge.puppet.com/puppet/r10k

[jcpunk]

To try and limp towards r10k support, I've opened #853. It is not a perfect solution, but it gets what I need out of the integration.