Project Infrastructure Migration 2007
This document attempts to discuss the needs of projects currently going through incubation. It attempts to address infrastructure needs, migration strategies to OSGeo servers, and migration strategies to mitigate disruption if OSGeo stops using CollabNet services.
Infrastructure Components
The following components make up the typical needs of an open source project, each section documents what the current projects use, and what CollabNet offers. The "pain" of migration is evaluated by Frank (thanks Frank).
Web Pages
Currently the members projects use:
- Wiki: OSSIM + GRASS (Twiki), GeoTools + MapBuilder (Confluence), Mapbender + OSGeo (Mediawiki)
- CMS: MapServer (Plone)
- Static HTML: MapGuide, OSGeo
- PHP HTML: GRASS (but CMS forthcoming) + weekly generated software user docs
- Doxygen (nightly generated HTML): GDAL, GRASS (for programmer's manual)
CollabNet offers static HTML pages under SVN for web sites. This works fine for MapGuide and GDAL. Not acceptable to MapServer nor GRASS. Migration is acceptable for Mapbender as most action is taking place in the current Wiki (and this will not change). Not sure about OSSIM, MapBuilder or GeoTools.
It would seem that a Wiki solution for the web site should be offered.
For MapGuide and GDAL migrating into the CollabNet website mechanism is no problem. Migrating out again (if needed) should also be straight forward. Migrating in and out for other projects would be moderately difficult due to all the reformatting needed.
From Daniel Brookshier - We can put any collaboration tool, like we did for the OSGeo Wiki as a subdomain of OSGeo.org. We can also do a cookie-based single signon. If you decide to move and can't live without such a tool, please consider moving your code repository and putting other resources as subdomains to osgeo.org. I am happy to oblige as wiki capability in the CollabNet tool is many months down the road.
Source Code Control
Some projects now using SVN, while others use CVS.
CollabNet offers both. From Daniel Brookshier -Note, we only have svn turned on for requests at this time. If you really want cvs, let me know and I can configure it for you.
Migrating in should be quite easy. The main downside is that all existing committer authentication will be lost, and will need to be resetup via the CollabNet infrastructure.
Will the CVS/SVN history be maintained when moving into/out of CollabNet? From Daniel Brookshier - There are tools like cvs2svn - not sure about how complete it is, but you can check out the docs and see if it meets your needs. CollabNet is also about to launch services to do this where our people do the import/conversion from cvs to subversion.
Some projects currently use CVS (or SVN?) triggers to launch actions such as IRC (via CIA-bot) notifications, mailing list notifications, web site updates, and automated builds. Collabnet provides a mailing list for updates, but does not allow arbitrary commit hooks (as far as I know). Perhaps we should look at using the commit mailing list to drive stuff like CIA. From Daniel Brookshier -The only hook we currently support is to a standard list in each project. If you submit a specific request t me detailing the hook and documentation to do it with a standard svn, I will put in a statement of work to CollabNet to request it. As with all changes to the underlying software, I can't promise anything until we run the request through to its conclusion.
CVS projects might want to take this opportunity to consider SVN which is superior technology. Howard Butler is knowledgable about how to do a CVS to SVN transition that preserves history. This would of course add some extra disruption for developers. From Daniel Brookshier - one key feature is https access with is far better than ssh tunneling for getting around your corporate firewall (note that in some companies, opening a tunnel is a fireable offense and in some countries you can be sent to prison).
Migrating out of Collabnet SVN is pretty easy assuming Collabnet provides access to the raw SVN archive (which they have agreed to do). The main disruption would be related to user authentication and a new location. SVN is open source so there is no need to change to a new tool if migrating out. From Daniel Brookshier - yes we support a full svn image in the event of migration out.
Bug / Issue Tracking
Currently members projects use:
- Bugzilla: GDAL, MapServer, OSSIM
- Jira: GeoTools, MapBuilder
- SF bug tracker: Mapbender
- RT: GRASS (Gforge planned)
- CN Issue Tracker: MapGuide
CollabNet offers an issue tracker with roughly comparible capabities to other bug trackers. (please note significant distinctions here) However, it is not clear that there is a clean way to migrate bugs from other systems to the CN issue tracker. Lacking this it seems unlikely the projects with substantial historical bug databases will be eager to migrate to the CN infrastructure.
Should any projects migrate to the CN issue tracker, it is also unclear that there is any way to migrate bugs out again should we drop CN support, though apparently CN does support an XML export of the database so hand crafting tools should be *possible* with some lossiness.
It would seem that migration to (and from) the CN bug tracker are going to be painful.
Mailing Lists / Forums
Currently member projects use:
- Mailman: GDAL, GRASS, OSSIM, MapBuilder, GeoTools, MapServer, Mapbender is currently in transition
- CN Mailing Lists: MapGuide, Mapbender
CN offers a mailing list mechanism (ezmlm). It supports and easy mechanism for administrators to batch subscribe email addresses, so migrating existing lists to it is relative easy (though digest or other config options may be lost). It has a few quirks (no apparent way to limit messages by size, digest is by accumulated mail size rather than something like daily). But the mailing lists seem to work fine, and are integrated into the platform. Migration will require all subscribers updating their address books with a new email address. The Mapbender PSC has experimentally copied the existing dev-mailman list to the CN ezmlm box. Works fine, no interruption. This is not solution regarding the archive which is left behind.
However, is it possible to migrate the mailing list archives when moving into/out of CollabNet? Some projects such as GRASS and GDAL do want to maintain it.
Migrating out it is easy to capture the subscriber lists, and setup a new external mailing list instance. It may involve a new email address again and will likely result in loss of email subscriber options. Is there an option to get the mailing list archives out of CN?
Overall, there is not a high cost to migrating in or out of the CN mailing list architecture.
The above does not address migrating archives which is also high desirable, but apparently very difficult with the CN mailing list archive manager.
Download Server
Existing projects offer source, binary and data downloads through http and ftp.
Collabnet offers an http based download facility from from the "Documents and Files" area on the left nav bar. This seems roughly analogous to the download support in SourceForge, and is generally adequate for downloads. (are there any perceived problems?)
There may be a migration hassle for folks moving large amounts of existing files into CN but generally speaking migration to CN for downloads should be straightforward.
Migrating out should also be relatively easy. Just "wget" the files to another server or something similar.
One downside of CN is that it doesn't offer ftp download services, but it isn't at all obvious to me that this is important in 2006. (comments?)
The other issue that could arise is that sufficient popularity for OSGeo projects could push the limits of our CN bandwidth limit (not sure what it is) in which case we might need to move some big things to the telascience hosted servers.
Wiki
Current projects:
- No wiki: GDAL (want one!), MapServer (had one but wiki-spammed), MapGuide
- Twiki: OSSIM, GRASS
- Mediawiki: Mapbender, OSGeo
- Confluence: GeoTools, MapBuilder
Currently CN does not offer a wiki. Arnulf has kindly asked terrestris (Till Adams) to host a Mediawiki instance for OSGeo.
It isn't clear if there are benefits to moving into a common wiki for projects. We could likely host Mediawiki, Confluence, Twiki, etc. instances for each project at telascience if needed.
There would be no pressing need to migrate out as the Wiki won't be dependent on collabnet.
Automated Build/Smoke Test System
Current projects:
- GDAL: BuildBot (prototype)
- GeoTools: cruise control + maven 2 (and may consider Continum)
- GRASS: script based build system for Linux, MacOSX, mingW; script/HTML based testsuite; additionally automated clone testing and function size/structure Quality Control
- Mapbender: currently testing Selenium
- MapBuilder: Setting up a fitnesse/ruby/WATIR/Autoit solution for AJAX type testing
- MapGuide: Cruise Control (I think)
- MapServer: BuildBot (prototype)
- OSSIM:
Moving automated Build/Smoke Test Systems to CollabNet infrastructure could go along with a build farm. Instead automated Build/Smoke Test Systems could also be installed on the telescience hosts.
Demo Site
Current Projects:
- geotools: n/a library project, demo examples in wiki and included with source download
- GRASS: a couple of Demo Live CDROMs/DVDs are available (Linux, MS-Windows)
- Mapbender: There are several Mapbender demo installations operated by different companies and clients, all linked from the Wiki. It is planned to create a "full stack" demo site on the telescience servers asap. Every Mapbender installation contains demo data (Capabilities URLs) and demo interfaces. A MapServer WMS demo ist operated by CCGIS hosting the Free Data project "Germany", a GeoServer WFS-T demo installation hosts the Mapbender users.
IRC
- GDAL: irc://irc.freenode.org#gdal
- GeoTools: irc://irc.freenode.org#geotools
- GRASS: irc://irc.freenode.org#grass
- Mapserver: irc://irc.freenode.org#mapserver
- Mapbuilder: irc://irc.freenode.org#mapbuilder
It would be desirable to have an automated archieving of IRC channels.
Security
- Common LDAP infrastructure for single sign-on
- SSL certificates for OSGeo sites (currently CN owned ?)
Further integration
Hold off on these until everything else is up and stabilized (and DNS is migrated).
Infrastructure Integration
This section describes the ways in which the components of the infrastructure interact, and this is where CollabNet starts to excel, however some catching up is needed to match the GeoTools project (Jive).
Collabnet Integration
| SVN | Build | Bug | IRC | Wiki | Web | Download | ||
|---|---|---|---|---|---|---|---|---|
| SVN | 1 | no | yes | yes | ||||
| Build | ||||||||
| Bug | ||||||||
| IRC | ||||||||
| Wiki | ||||||||
| Web | ||||||||
| Download |
1. Can use an existing system such as curise control here
Q: Does the Collabnet Issue tracker interact with anything? Can we get an email of new bugs? Reply to those messages to comment?
Okay I am not doing Collabnet any justice here, what does it integrate? And how... Jive 14:29, 31 March 2006 (CEST) Can someone fill in the above table Jive 14:21, 31 March 2006 (CEST)
GeoTools Integration
Here is a worked example illustrating how the GeoTools project is intergrated:
| SVN | Build | Bug | IRC | Wiki | Web | ||
|---|---|---|---|---|---|---|---|
| SVN | yes | 1 | yes | yes | yes | ||
| Build | yes | yes | |||||
| Bug | yes | yes | yes | ||||
| IRC | 2 | ||||||
| yes | |||||||
| Wiki | |||||||
| Web | yes | yes | yes | yes |
- not sure if this one works right now
- logs of IRC meetings are posted
Areas for improvements in integration:
- single user name / password for svn, confluence, jira
GeoTools SVN Integration
We are capturing integration in a single direction: ie if we do a commit in svn what gets updated? Well everything in the following list...
- SVN (commit) to Build
- curise control builds after a 30 min pause in commits
- curise control builds nightly
- various cruise control instances watch different important branches, these are maintained by several organizations
- SVN (commit) to Tracker
- the Jira tracker will watch the svn code repository and pick up any commit message that mentions a Jira Number, these show up as comments against the Jira Issues, and serve to document a patch being applied in development and then stable branches
- (not sure if we have configured this correctly)
- SVN (commit) to IRC
- CIA is used to host a bot on the freenode#geotools IRC channel it cheerfully chrips up about commits, but does not answer questions
- SVN (commit) to Email
- there is a geotools-commits email list that can be subscribed to
- SVN (commit) to WIKI
- you can use wiki syntax to grab lines of a file from svn, these can be used as code examples on demo and tutorial pages
- SVN (commit) to Web
- cruise control is used to generate a javadoc website based on svn
- we build a website for module information based on maven 2, a series of performance metrics are gathered using tools like clover.
Tool Selection Criteria
If a project were to leave OSGeo, then they should be able to set up and use all the tools without a license cost.
- Ideally, the tools should be open source, but a "Free for Open Source" licence is acceptable.
- Tools that use Open Standards will be used where possible as reduces the need for vendor lock in.
- Tools should be able to input data from existing OSGeo projects with minimal effort and without loosing history.