On 01/29/2018 10:02 PM, Bryan Turner wrote: > On Mon, Jan 29, 2018 at 11:08 AM, H <agents@xxxxxxxxxxxxxx> wrote: >> I am a newcomer to git looking to set up a web development environment where individual computers are used for development, the development.git, staging.git and production.git repositories are stored on an external server reachable by password-less ssh and the staging and production websites are on yet another server, also reachable by password-less ssh from the git-server (and the development machines). >> >> Locating the three git repositories on an external server works fine but I have not been able to have the staging and production deployment files on another server. I believe this is what is referred by GIT_WORK_TREE and based on what I found on the web I created a post-receive hook of staging.git with the two lines: >> >> #!/bin/sh >> GIT_WORK_TREE=user@1.2.3.4:/var/www/html/dev.whatever git checkout -f master >> >> I believe this should deploy the files from the development work tree. >> >> The above, however, fails. Should it work? I am running git 1.7.1 on CentOS 6. > No, I wouldn't expect that to work. GIT_WORK_TREE is not remote-aware > in that way. It's expected to be a normal-ish filesystem path. > > Based on your description, and the hook you've written, it seems like > your intention is for the source to automatically be fetched and > checked out on the staging environment after each push. (This is > dangerous, and likely _not_ what you actually want, but I'll get to > that in a moment.) > > One option would be to setup something like NFS, so the git-server can > mount the filesystems from the staging and production nodes. > > A different, likely better, option would be to have the post-receive > script on the git-server use straight ssh to trigger a checkout script > on the staging server, e.g.: > #!/bin/sh > ssh example@staging-server -C /opt/deploy-staging.sh > > Your deploy-staging script would then do something like: > #!/bin/sh > GIT_WORK_TREE=/var/www/html/dev.whatever git pull origin > > That said, though, having such a simple script is dangerous because > Git is fully capable of having receiving multiple pushes concurrently, > and they can all succeed as long as they're updating different > branches. Since your script isn't considering what branches were > changed by the push, it could end up triggering simultaneous git > processes on the staging server all attempting to deploy concurrently. > > The stdin for the post-receive hook receives details about which refs > were changed, and you'll likely want to update your script to parse > stdin and only try to deploy staging if a specific, relevant branch > (master in your example) has changed. > > Lastly, I'll note that using post-receive will make the pushing > (remote) user wait while the staging server is deployed. If that > process is likely to take very long, you might want to decouple the > two somehow. > > Hope this helps! I should perhaps also have mentioned that although I am the only developer, I may use different computers to develop on. IOW, there should not be any conflict due to code being pushed by multiple developers. Let's see if I understand this correctly: - Unless NFS is used, the git archive and the deployment of the website code in this case should reside on the same computer. - The combination of the checkout script and the deploy-staging script should work provided not multiple updates to the same branch are pushed at the same time. I will try this later today but any other hints or suggestions you may have would be greatly appreciated!
