I would like repeatable results when running maven commands locally, even if somebody else is pushing updates to a snapshot dependency.
To achieve this, I would like to use the updatePolicy of never.
This will allow any dependencies that aren't available locally to be downloaded, while any I have installed locally will be used.
The offline flag won't work in this situation, as there may be dependencies that I haven't installed locally which will need to be downloaded from the remote repo.
I don't want to have to modify the pom, as doing this locally with every checkout will be error prone, and I don't want to commit these changes as it will have adverse effects on other developers.
Ideally I'd like to specify this from the command line. The opposite of the -U flag.
I've searched the docs, and so far have not found out how to do this.
If you want repeatable builds you can create a Docker image that can run Maven. Then load all you project files and run Maven build.
This will provide a clean environment for your build every time.
About the changing dependencies, if you work using SNAPSHOT dependencies, you must expect this different results. That is what SNAPSHOT means: "this is under development".
If you (or your team) control the SNAPSHOT dependency and there is an error in the build that's a "good" sign, the tests found something to be fixed.
If you (or your team) don't control the SNAPSHOT dependency, you would prefer to the last stable release.
Related
For various reasons (mostly speed and reliability* of build) I prefer to use maven in offline mode. As I don't like having to type -o every time (and deal with problems when I forget it), I've added <offline>true</offline> to my .m2/settings.xml.
However, sometimes I have to build online. Is there a command-line option for this, or any decent way of doing it without having to edit settings.xml and restoring it after the build?
* Why offline helps reliability in my case - the project consists of several maven projects, each with a set of submodules, so it requires running seveal mvn commands in sequence, each taking quite some time to build.. As all of those are automatically built on Jenkins, without offline sometimes later builds contact company maven repository and fetches artifacts that were built just during my local build, and I end up with a local build that does not contain my changes.
Regardless of build tool, I have seen people doing a clean task/phase before every time they do package/compile or ... is it really necessary?
Dose build tools use reuse artifacts of previous builds?
Most of the time you see clean install as the default command, but I would encourage everybody to use verify instead.
When executing clean the target-folder is removed which makes it impossible to do incremental builds. Plugins have enough information to detect if they should do their action. For instance: the maven-compiler-plugin compares the java sourcesfiles and the compiled classfiles (and other things) to see if files needs to be (re)compiled. If you think that a plugin is not working correctly with incremental builds, please file an issue for that plugin.
The install was often required with multimodules in Maven2, but Maven3 is capable to resolve these inner module dependency references. The only thing 'install' does is copying artifacts to the local repository (=IO=expensive). And it'll make your local repo look different compared to your coworkers, which might give different results during builds. Better to let a buildserver push those artifacts to the shared remote repositories and let every pull those SNAPSHOTs from there. Only in rare cases calling install is valid (experienced Maven users know when :) ), so instead please use verify.
I have Jenkins version 2.7.1 running on a Windows 7 machine. It is successfully pulling code from a subversion repository and running tests. I have the test jobs set up for the development branch of each project only.
We periodically make stable releases of the projects in jar files with version numbers. I would like to have Jenkins be the repository manager for those stable releases. These are made by hand - There is no Jenkins job making or testing stable releases. The projects do use Maven.
Each stable build is tagged in the subversion repository, so it could be made again on demand if needed.
I downloaded the Maven repository server hoping to make this fit the purpose. I read the documentation that's provided, but it's pretty terse. As I understand it and have it configured now, this appears to have a couple of issues:
If I go to jenkins-ip/plugin/repository/project, it has made directories there that expose the names of all of my projects, which seems undesirable. (Here jenkins-ip is the IP where I access Jenkins on my local network.)
On the other hand, there's nothing but empty directories under these projects, so they're currently useless.
These projects all correspond to the continuous testing of the development branch. There's no apparent way to get the stable builds into the hierarchy. (It doesn't seem efficient to create a job for each stable release...)
Is there anyway to get Jenkins (with this plugin or through another method) to be the repository manager just for the stable builds? I know that I can start a different repository manager like archiva, but it would be ideal to use Jenkins since it's already running and it seems to claim capability for this function now.
To use Maven repository server you have to build the project on Jenkins.
Then the plugin will expose all archived artifacts as maven repo.
Note you need to use a "Maven project" type for it to work (freestyle is not supported)
There are several plugins that will help you manage building from multiple tags, however not all of them work with "Maven project" type.
You could also try Jenkins pipeline (previously "Workflow") or the Job-DSL plugin.
A simplest solution would be to have a build parameter specify the tag name (then checkout e.g. ^/tags/projectname/${tagParam}), but you have to figure out how to trigger the job then.
I have a Jenkins job that builds a simple Maven project. If all I do is build, it works just fine. The problem arises when I try and do a release, dry run or regular. It consistently fails with the Cannot prepare the release because you have local modifications error. I have wiped out the workspace, but the problem persists. Is there any way I can get Maven to tell me which file it thinks has been modified? I would assume that by wiping out the local workspace and immediately running the dry run release that there wouldn't be any opportunity for anything to get modified.
Please note, I do not have access to the Jenkins server or the slave that is running the actual release build, so I can't use any tools there (like SVN) to determine what is supposedly modified.
You can use the Maven SCM plugin to do a diff.
https://maven.apache.org/scm/maven-scm-plugin/diff-mojo.html
Basically, integrate the maven plugin upstream of the failure, and see if anything has been changed. I imagine you might be able to see the output in the log, but if you cannot, you might be able to move your "real" maven pom.xml aside and replace it with one that generates a diff file and with the help of the maven build helper plugin, attaches that file as an additional aritfact (to a pom target).
It turned out the solution to my problem was to not use the "Local to the workspace" strategy for my private Maven repository in the Jenkins job configuration. By changing that to the "Local to the executor" strategy the problem went away. I'm still not sure why it was having the problem in the workspace, but this solution resolved it form me, and might work for others.
Well, this is kind of embarrassing. I am in the process of mavenizing our build processes and just don't know how the access the result of a build. I build, let's say, a jar file and mvn deploy it. So it ends up as some blah-0.1.2.jar in our company maven repository, which is just a webdav share. Now how would you pass that on to someone else to use? Just pry it from target/blah-0.1.2.jar can't be the answer. I found several suggestions to use variants of mvn dependency:get but they were all just close and didn't feel right. There must be a way to use all those nice versions of blah-*.jar that end up in the repository for purposes other than a maven dependency. Preferably from the command line and maybe even without maven. Hm, a webdav client doesn't look too bad except for snapshots. What would you suggest?
Creating a script that makes a dependency:get call is probably going to be closest to your desired outcome. You can specify the destination of your downloaded jar this way.
If you are looking for an easy way to share builds between people in/outside of your company then you can look into setting up some automated build software like Bamboo or something similar. A new build gets triggered any time a commit is made to the section where your project resides in whatever version control system you use. An artifact is then made available for each successful build and are available via Bamboo's web interface. Bamboo can be configured to run with your maven pom's.
While they can bit a bit of pain to set up, going the automated build route will take a lot of the sting out of sharing your builds in the future.