Provide new 'cwd' option during deploy

[JoshuaGross]

I don't really understand how to use Naught in a robust way for new deploys. My current deploy script has a git post-receive hook that brings the old server down and a new server up; if the new server fails, it will rollback and bring the old server back up. One thing I'd like to do with Naught is provide new CWDs in a deploy and then periodically check the health of the new deploy and switch back to the old CWD if necessary. What I'll do for now is symlink a directory and point Naught to that, but this seems strange.

[andrewrk]

Your request makes sense; however I think we may be limited by node's cluster API.

[JoshuaGross]

Ahh, I didn't realize that. Makes sense. Do you manage production deploys in a similar way to me / do you have any suggestions for how to do this elegantly?

[andrewrk]

We currently use rodent to do production deploys. It may not be generic enough for your use case, but here's what a deploy looks like:

• developer pushes the code to a remote branch (usually master)
• developer executes the deploy command
• on each production target
  • git fetch
  • git checkout origin/(deploy-branch)
  • npm install
  • ./node_modules/.bin/naught deploy

So there are no backup folders with previous releases to roll back to. However, deploys are very fast, usually taking about 5 seconds. So if you want to roll back to a previous commit, you can deploy that previous commit and the previous commit will be live in 5 seconds.

[JoshuaGross]

I think I like that better than the current system, which takes much longer than 5s to deploy and doesn't really use the power of Git. Thanks - this is informative!

Joshua Gross Christian / SpanDeX, Inc. / BA of Computer Science, UW-Madison 2013 http://www.joshisgross.com / http://www.spandex.io

On 5 March 2013 14:07, Andrew Kelley [email protected] wrote:

We currently use rodent https://github.com/indabamusic/rodent to do production deploys. It may not be generic enough for your use case, but here's what a deploy looks like:

• developer pushes the code to a remote branch (usually master)
• developer executes the deploy command
• on each production target
  • git fetch
  • git checkout origin/(deploy-branch)
  • npm install
  • ./node_modules/.bin/naught deploy

So there are no backup folders with previous releases to roll back to. However, deploys are very fast, usually taking about 5 seconds. So if you want to roll back to a previous commit, you can deploy that previous commit and the previous commit will be live in 5 seconds.

— Reply to this email directly or view it on GitHubhttps://github.com/indabamusic/naught/issues/25#issuecomment-14462133 .

[iuriaranda]

We perform deploys the capistrano / chef way, so the deploy process creates a /release/<version> directory and then creates a symlink from /current to the latest deployed version dir. So we face the problem of the cwd in the deploy command.

A quick workaround that I've seen here: http://clarkdave.net/2013/02/node-js-and-cluster-cwd-dirname-shenanigans/ would be to chdir inside the main script of the app using the __dirname variable:

if (__dirname !== process.cwd()) {
  process.chdir(__dirname);
}

Even this does not involve any change in naught, I think it would be cleaner if there was a way to change the cwd during a naught deploy.

[andrewrk]

Feel free to make a PR. This feature can easily lead to subtle problems - I think we want an automated test or 2 for this feature.

[andrewrk]

This is now implemented and released in 1.2.0 thanks to dmcaulay

[danshulman]

I have a node.js app (express) in folder1. I start the server using "naught start app.js". I now would like to deploy the updated version located in folder2. I run (from folder1) "naught deploy --cwd ../folder2".

I get the following result:

SpawnNew. booting: 1, online: 1, dying: 0, new_online: 0 NewOnline. booting: 0, online: 1, dying: 0, new_online: 1 ShutdownOld. booting: 0, online: 0, dying: 1, new_online: 1 OldExit. booting: 0, online: 0, dying: 0, new_online: 1 done

The server is still up, but folder1 is still deployed, instead of folder2. I would like to make the switch from folder1 to folder2 with zero downtime. Any suggestions?

[andrewrk]

cc @dmcaulay

[dmcaulay]

@andrewrk I'll take a look

[dmcaulay]

@zenshulman I've updated my fork with a possible fix. I've tested locally, but I want to make sure it works for you before opening a PR. Can you test with my fork? https://github.com/prix-fixe/naught

[dmcaulay]

@zenshulman I want to clarify that the cwd option does not change the script location. I'm using a symlink for that.

For example if you run naught deploy --cwd '/release/2' it doesn't change the script file location from '/release/1/server.js' to '/release/2/server.js'. It only changes the cwd directory for the master process. This is needed because the workers inherit the cwd from the master process.

I think we could update the script file, but that would take a little more work. What do you think @andrewrk? Should the cwd option change the script location on deploy?

[danshulman]

Oh, I thought cwd role is to change the script. Using symlink works, but what's cwd for?

[andrewrk]

if cwd exists, it seems like it should work for this use case, right?

[danshulman]

For what use case? Switching scripts from different folders?

[andrewrk]

Yes. This is your use case, correct?

[danshulman]

It is. But using cwd alone with no symlink doesn't change the script.

[andrewrk]

Sorry, to be clear, I am arguing in favor of your use case. I am saying that as is it is broken and we should support this use case.

[danshulman]

ah, ok great. Yes, I 'm totally in favor of that :)

[dmcaulay]

i just reviewed the docs and im not sure if that's possible. check it out. http://nodejs.org/api/cluster.html#cluster_cluster_setupmaster_settings

"Only the first call to .setupMaster() has any effect, subsequent calls are ignored"

also cluster.settings http://nodejs.org/api/cluster.html#cluster_cluster_settings

"It is effectively frozen after being set, because .setupMaster() can only be called once."

[dmcaulay]

@zenshulman you need to change cwd because it's used when you require files. If it is not updated then require will load the files from the old directory.

[danshulman]

cool, got it!

[danshulman]

Wait, so how come when I deploy after changing only the symlink ("naught deploy" only, no pwd option), I see changes from the new folder?

[dmcaulay]

Because cwd defaults to the cwd of the naught script.

[andrewrk]

I looked into it, and sadly I think @dmcaulay is right - we already call process.chdir with the new cwd when you deploy, but it doesn't work for the script.

See https://github.com/joyent/node/issues/3893 and https://github.com/joyent/node/issues/6264

[danshulman]

ah ok, so I'll use the symlink method then.

[dmcaulay]

@andrewrk we should probably update the readme so this is clearer. do you want me to take a stab at it and open a pr?

[danshulman]

sure, why not

[andrewrk]

@dmcaulay yes please. meanwhile I opened the above issue in joyent/node

[freezy]

Great! Started testing this out yesterday and didn't work, then I read through the GitHub issues here, saw it was fixed three hours before, pulled from repo and it started working. :)

I've described how to setup naught with an upstart service wrapper and Git push deployment here in case anyone is interested. All the rest is still WIP though.