The way I solved this (and a host of other issues) is by specifying that the ivy directory be in the Hudson project workspace (I actually started the practice with CruiseControl). It is a less efficient at build time because the cache is isolated for each build, but the trade off is that I don't have to worry about any concuurency issues in the cache, nor any side effects between builds. I can't get at my Hudson server right at the moment, but if you need to know what variable I set, please ask.
Josh On Monday, August 31, 2009, Willem Verstraeten <[email protected]> wrote: > I'm using the Hudson CI server. > > The thing is that, if I don't set the resolveId, I can have two > projects which are performing the exact same resolve. > > Ie. project A is retrieving, inline, the '[com.mycorp]-[common]' > module, and project B as well, possibly at the same time. Now the > resolveId for both these resolves is the [com.mycorp]-[common-caller], > as they are both inline resolves. Since they have the same resolveId, > they also use the same files in the ~/.ivy2/cache directory, which > leads to errors when they both run at the same time. > > So I'd like to set a resolveId that is dependent on the > http://ant.project.name. I'd also like the resolveId to depend on the ivy > branch, as, again, it is possible that the same project is built for > two branches simultaneously, resulting in the same resolveId again if > the branch is not part of the resolveId. > > So, in short, is there a way to specify in the ant file a resolveId > for an inline retrieve that depends on the dynamic Ivy properties such > as org, rev and branch ? > > If not, is there an alternative way in which I can prevent the > concurrency issues described above ? > > I hope this makes my problem a bit clearer. > >> Joshua Tarp wrote: >> >> What CI build server are you using? I've never heard of cross-Ant >> concurrency issues that didn't involve the filesystem. I thought most >> if not all forked Ant. > >> On Friday, August 28, 2009, Willem Verstraeten >> <[email protected]> wrote: >> > Hello, >> > >> > How can you set a resolveId that depends on the properties of the >> > module to be resolved ? >> > >> > Ie. I am trying to do something like this: >> > >> > >> > <ivy:retrieve inline="true" >> > organisation="com.mycorp" >> > module="common" >> > type="buildscript" >> > conf="build" >> > >> > resolveid="[org]-[module]-[branch]-${http://ant.project.name}";; >> > pattern="${common.dir}/[artifact].[ext]"/> >> > >> > The resolveId is not treated as a pattern though, so the resolveId is >> > the literal [org]-[module]-[branch]-project_name, iso >> > com.mycorp-common-development-project_name. >> > >> > Now for the [org] and [module] parts I can obviously fill them in as a >> > literal, because I specify them as literals anyway. However, the >> > branch is determined by the settings file, so I'd like to avoid >> > duplicating that logic here. >> > >> > Also, when I would not do an inline resolve (ie. resolve based on an >> > ivy file) I guess I would need this functionality as well, as I don't >> > have the literal information available then. >> > >> > All this would be necessary to support concurrent building of several >> > modules which all start with this piece. Ie. I have a multi-module >> > setup, and the build scripts all use the above retrieve task to get >> > the common parts of the build script. If two modules are being built >> > at the same time (which is quite likely when you have a continuous >> > integration server), you get the possibility that two build scripts >> > are resolving the same module at the same time with the same >> > resolveId, leading to concurrency problems. >> >
