3

I have a website that I want to deploy to a clients DEV and UAT environments, the site is part of a mercurial repo - it is in the Website folder at the same level as the .hg folder. I know I can push the entire repository but would rather push only the website folder so the client does not have the other files and folders.

The repo looks like this:

  • Project root
    • .hg
    • Database (SQL Source Control uses this)
    • Documentation (All specs, pdfs, art work etc.)
    • Lib (pre-Nuget 3rd party dlls)
    • packages (Nuget stuff)
    • Website (this is the only area I want to deploy)
    • .hgignore
    • Project.sln

Edit: The clients servers are not connected directly to the internet, my access to them is over a vpn and then RDP. Currently to deploy any changes I need to zip the site up, put it on a shared ftp server then wait up to 3 days for the files to be copied to the servers. Rules have been configured so I can use Mercurial over this connection.

Edit 2 I have managed to create a subrepo from the Website folder by forgetting the Website folder and all it's contents, committing the change then putting the files back, creating a repo then echoing out the .hgsub file. Locally this works for me, I can clone from the Website repo without getting any of the additional folders. However I have not been able to use this version of the repo, even if I repeat the process on our repo server. When I try to clone the hosted version down to my local working copy I get 404 errors, but I can clone the hosted version on the hosting server.

I would appreciate some step-by-step instructions (a guide for dummies if you like) on how to achive my goal; which is to be able to push only the Website folder to the clients servers. The master copy of the repo is on our repo server, I have a local clone and need to be able to push out versions from my copy.

Edit 3 Turns out that the problem I was having converting a folder to a subrepo as described in http://mercurial.aragost.com/kick-start/en/subrepositories/#converting-folder-into-a-subrepository was that the convert command, in versions after 2.1.0, is broken and is still broken in 2.3.1. After I figured that out and rolled back to that version of TortoiseHg I was able to convert the folder to a subrepo, in the root of the repo I have .hgsub which says Website = Website. I was able to work with that locally, commit to the whole repo, the subrepo, clone either the full repo or the subrepo (which is what I want), however I can't get this to work from our master repo server.

I zipped the whole thing up and ftp'd it to our remote master repo server, then set it up so I could clone from it. Directly on the server this works fine (hg clone --verbose -- C:\Repositories\EM .), however when I try to clone from the server to my local development machine with (hg clone --verbose -- https://myserver.com/hg/EM/ .) it fails with "HTTP Error: 404 (Not Found)".

requesting all changes
adding changesets
adding manifests
adding file changes
added 628 changesets with 6002 changes to 4326 files
updating to branch default
resolving manifests
calling hook preupdate.eol: <function preupdate at 0x00000000035204A8>
getting .hgignore
getting .hgsub
getting .hgsubstate
HTTP Error: 404 (Not Found)
[command returned code 255 Fri Apr 20 10:51:23 2012]

I don't know what the problem is, the files are there so why the 404?

Simon Martin
  • 4,203
  • 7
  • 56
  • 93

4 Answers4

2

In my opinion Mercurial shouldn't be used for this purpose. This is particularly true if that website is a web application because you shouldn't have the DLLs in Mercurial.

You should look at the web deployment tool built into Visual Studio. Have a look at this page to see if it suits your purpose.

If you can't install the required services on the destination server then it can be configured to use FTP instead.

Steve Kaye
  • 6,262
  • 2
  • 23
  • 27
  • I can't install onto the destination servers and they don't have direct internet access. Currently I have to zip files, ftp up to a server then wait 3 days for them to be copied to the machine. Using Mercurial I can deploy directly over the vpn I have. Also it means there's some sort of versioning going on. – Simon Martin Mar 23 '12 at 10:09
1
  1. You can not push part of repo tree
  2. If DEV and UAT environments are unversioned targets, you can use any other way for distributing Mercurial content
  3. You can separate Website into subrepo and will be able to push this repo
Lazy Badger
  • 94,711
  • 9
  • 78
  • 110
  • Working with a subrepo is new to me, but I've created a subrepo from the Website folder and can push that as you say. Do you know any way to have the Workbench always supply the --subrepos switch? I can do it from commandline but it's not my normal workflow – Simon Martin Mar 23 '12 at 10:21
  • @SimonMartin - dirty hack (TBT!), but - you *can create aliases* in hgrc of repo, redefining needed commands under the same name, but with added options – Lazy Badger Mar 23 '12 at 10:36
  • With the subrepo I can't push back up to our central / master repo. I tried creating a "deploy" copy, cloned my main local repo and set the subrepo in there, then cloned only the sub repo which worked. However I couldn't then update my "deploy" repo from my main local repo. I just edited web.config to test and the error is "abort: path 'Website\web.config' is inside nested repo 'Website'. While I _can_ create a deploy clone, create the sub repo then use that to push to the client it is a bit long winded. How can I avoid the need to create a deploy clone each time? – Simon Martin Mar 23 '12 at 11:19
  • @SimonMartin - "abort: path 'Website\web.config' is inside nested repo 'Website' **means exactly same thing as written** - you still have **nested repo instead of subrepo**. Convert nested into subrepo - read MercurialWiki how to do it, **I don't want and do not quote it here** – Lazy Badger Mar 24 '12 at 05:47
0

I have a working solution to this. I created a batch file that creates an outgoing repo and starts the built in server so I can pull from it on the client machines. First it clears out the previous folder, then clones from my local working copy (there's a parameter to determine which tag it should clone from). Next it creates a map file and converts the Website folder to a new Website2 folder in order to preserve the history then gets rid of the original folder and renames the new one. Finally it spins up the built in server.

cd c:\inetpub\wwwroot
rd /S /Q _ProjectName
hg clone -- C:\inetpub\wwwroot\ProjectName#%1 C:\inetpub\wwwroot\_ProjectName
cd c:\inetpub\wwwroot\_ProjectName
echo include Website > map.txt
echo rename Website . >> map.txt
hg --config extensions.hgext.convert= convert --filemap map.txt . Website2
cd Website2
hg update
cd ..
hg remove Website/*
hg commit -m "Removed Website"
rename Website2 Website
hg serve

So it isn't pretty, but now I just need to call the batch file and pass the tag I want to build the outgoing website from (uat, dev etc.) and give it a minute to create my Website folder, with history, that I can use to pull from or push from. I don't need to call hg serve because I know the names of the client servers so I can push the changeset out by creating aliased remote repositories. But I included that step so the client machines can pull. I haven't fully explored this option, so I'm not sure whether it's got any particular advantage. It's fine for the case when it's just me working on the project, but if any other developer needs to work on this then the Uri for their local project server will obviously be different (http://SIMON-PC:8000/ won't be the case for everyone), in which case pushing into the client might be best.

But by using this approach my local working repo doesn't need to change and so I don't get any issues communicating with our central repo, the 404 errors mentioned in edit3. I keep the entire history of the repo with the convert process, so the next time I need to send changes I'm not starting at revision 1 - in other words it isn't destructive of the Website and although I am deleting the entire outgoing repo (_ProjectName) each time I am retaining the history and yet in a position to pull / push ONLY the Website directory because it is created each time as a 'standalone' repo

Simon Martin
  • 4,203
  • 7
  • 56
  • 93
0

As others have pointed out you can't use push for this. Just do 'rsync' from your server to theirs. You could even automated that in a hook, where you push to a local repository and it auto-deploys to their site. Something like:

[hooks]
changegroup.deploy = $HG update ; rsync Website account@theirserver:/path/to/docroot
Ry4an Brase
  • 78,112
  • 7
  • 148
  • 169
  • Because you don't have to specify the revision any time you call this script. Hg itself says: "Update the repository's working directory to the specified changeset. If no changeset is specified, update to the tip of the current named branch." – Christoph Jüngling Mar 23 '12 at 09:48
  • @ChristophJüngling - `hg export -r tip` always. Faster, more secure – Lazy Badger Mar 23 '12 at 10:13
  • @LazyBadger Did you mean `hg archive` instead of `hg export`? `hg export` just gets you the tip patch, but I thought the goal was a snapshot of tip. I first wrote it with `archive` but you have to export that to somewhere and then you need to worry about collisions and locking. Updating on the non-webserver box and then rsyncing to the web server is just as secure. – Ry4an Brase Mar 23 '12 at 18:37
  • @Ry4an - mea culpa, hg archive, yes... I like archive more because: no authorization additiinally, I can copy to archive only modified files for single revision or revision-set and for huge repos **it does matter** – Lazy Badger Mar 24 '12 at 04:09