Programming, contributing: Using Maven with OSGi Part 2

Introduction

In previous installment I've described the foundation block of Maven - Eclipse Aether library. I've mentioned declarative usage of Maven using mvn: URI scheme.

In Karaf 4:

karaf@root()> bundle:install mvn:commons-io/commons-io/2.5
Bundle ID: 52

or in JBoss Fuse:

JBossFuse:karaf@root> osgi:install mvn:commons-io/commons-io/2.5
Bundle ID: 295

In this part I'll describe how Maven and Aether library are used in Karaf 4 and in JBoss Fuse 6.3.x (standalone mode). I'll leave JBoss Fuse fabric mode for next installment.

Let's recall some fundamental concepts:

local repository - accessed by Aether with the help of org.eclipse.aether.repository.LocalRepositoryManager interface and org.eclipse.aether.repository.LocalRepository class. Effectively local repository is a wrapper for locally accessible filesystem directory that follows specific structure (organization of Maven artifacts).

remote repository - accessed by Aether with the help of org.eclipse.aether.repository.RemoteRepository interface. Effectively remote repository is a wrapper for URI, a set of policies related to snapshot/release versions plus proxy, mirroring and authentication information.

pax-url-aether

The above commands rely on mvn: protocol implementation of java.net.URLStreamHandler interface. This implementation is provided by OPS4J PAX URL Aether library and uses Eclipse Aether to handle Maven artifact resolution.

When pax-url-aether bundle is installed and active in OSGi framework, it provides OSGi service with org.ops4j.pax.url.mvn.MavenResolver interface.

In Karaf 4:

karaf@root()> bundle:services -p 4

OPS4J Pax Url - aether: (4) provides:
-------------------------------------
objectClass = [org.osgi.service.cm.ManagedService]
service.bundleid = 4
service.id = 9
service.pid = org.ops4j.pax.url.mvn
service.scope = singleton
----
objectClass = [org.osgi.service.url.URLStreamHandlerService]
service.bundleid = 4
service.id = 10
service.scope = singleton
url.handler.protocol = mvn
----
objectClass = [org.ops4j.pax.url.mvn.MavenResolver]
service.bundleid = 4
service.id = 19
service.scope = singleton

In JBoss Fuse:

JBossFuse:karaf@root> ls 4
You are about to access system bundle 4.  Do you wish to continue (yes/no): yes

OPS4J Pax Url - aether: (4) provides:
-------------------------------------
objectClass = [org.osgi.service.cm.ManagedService]
service.id = 8
service.pid = org.ops4j.pax.url.mvn
----
objectClass = [org.osgi.service.url.URLStreamHandlerService]
service.id = 9
url.handler.protocol = mvn
----
objectClass = [org.ops4j.pax.url.mvn.MavenResolver]
service.id = 21

Configuration

This time I won't show any code examples. We don't need them. As with most of other OSGi services we can configure how pax-url-aether uses Aether with Configuration Admin service. Describing Configuration Admin is a good task for another blog post. There are some articles about Configuration Admin and if something's not clear, please refer to original documentation.

To configure org.ops4j.pax.url.mvn.MavenResolver service (the implementation is org.ops4j.pax.url.mvn.internal.AetherBasedResolver) we use org.ops4j.pax.url.mvn PID (persistent identifier). We can configure this PID manually using Configuration Admin API or using etc/org.ops4j.pax.url.mvn.cfg file.

Internally, pax-url-aether uses org.ops4j.pax.url.mvn.internal.config.MavenConfigurationImpl object. It contains information used by org.ops4j.pax.url.mvn.internal.AetherBasedResolver.

Let's see what's the default configuration (sightly formatted) provided by Karaf 4 (/work directory is a docker mount volume):

karaf@root()> property-list --pid org.ops4j.pax.url.mvn
   felix.fileinstall.filename = file:/work/etc/org.ops4j.pax.url.mvn.cfg
   org.ops4j.pax.url.mvn.defaultRepositories = \
      file:/work/system@id=system.repository@snapshots, \
      file:/work/data/kar@id=kar.repository@multi@snapshots, \
      file:/work/system@id=child.system.repository@snapshots
   org.ops4j.pax.url.mvn.repositories = \
      http://repo1.maven.org/maven2@id=central, \
      http://repository.springsource.com/maven/bundles/release@id=spring.ebr.release, \
      http://repository.springsource.com/maven/bundles/external@id=spring.ebr.external, \
      http://zodiac.springsource.com/maven/bundles/release@id=gemini, \
      http://repository.apache.org/content/groups/snapshots-group@id=apache@snapshots@noreleases, \
      https://oss.sonatype.org/content/repositories/snapshots@id=sonatype.snapshots.deploy@snapshots@noreleases, \
      https://oss.sonatype.org/content/repositories/ops4j-snapshots@id=ops4j.sonatype.snapshots.deploy@snapshots@noreleases, \
      http://repository.springsource.com/maven/bundles/external@id=spring-ebr-repository@snapshots@noreleases
   org.ops4j.pax.url.mvn.useFallbackRepositories = false
   service.pid = org.ops4j.pax.url.mvn

And JBoss Fuse:

JBossFuse:karaf@root> config:proplist --pid org.ops4j.pax.url.mvn
   felix.fileinstall.filename = file:/data/servers/jboss-fuse-6.3.0.redhat-145/etc/org.ops4j.pax.url.mvn.cfg
   org.ops4j.pax.url.mvn.defaultRepositories = \
      file:/data/servers/jboss-fuse-6.3.0.redhat-145/system@snapshots@id=karaf.system,\
      file:/home/ggrzybek/.m2/repository@snapshots@id=local,\
      file:/data/servers/jboss-fuse-6.3.0.redhat-145/local-repo@snapshots@id=karaf.local-repo,\
      file:/data/servers/jboss-fuse-6.3.0.redhat-145/system@snapshots@id=child.karaf.system
   org.ops4j.pax.url.mvn.globalChecksumPolicy = warn
   org.ops4j.pax.url.mvn.globalUpdatePolicy = daily
   org.ops4j.pax.url.mvn.localRepository = /data/servers/jboss-fuse-6.3.0.redhat-145/data/repository
   org.ops4j.pax.url.mvn.repositories = \
      http://repo1.maven.org/maven2@id=maven.central.repo, \
      https://maven.repository.redhat.com/ga@id=redhat.ga.repo, \
      https://maven.repository.redhat.com/earlyaccess/all@id=redhat.ea.repo, \
      https://repository.jboss.org/nexus/content/groups/ea@id=fuseearlyaccess
   org.ops4j.pax.url.mvn.settings = /data/servers/jboss-fuse-6.3.0.redhat-145/etc/maven-settings.xml
   org.ops4j.pax.url.mvn.useFallbackRepositories = false
   service.pid = org.ops4j.pax.url.mvn

The above configurations differ a bit. JBoss Fuse provides more explicit configuration. Let's describe each used (and assumed) properties that can be used to configure pax-url-aether (org.ops4j.pax.url.mvn.internal.AetherBasedResolver).

org.ops4j.pax.url.mvn.defaultRepositories

this is a list of local repositories searched for an artifact in the first phase of artifact resolution. This repository should not contain URIs other than file://-based. Each repository from this list is treated as local repository. pax-url-aether iterates over this list and checks one repository at a time. If neither location contains the artifact being resolved, pax-url-aether switches to second phase that involves remote repositories.
Also these repositories do not require write access - Aether doesn't write any files there.

Access to these repositories can be presented using the following code (did I promise not to show any code example? sorry...):

RepositorySystem system = locator.getService(RepositorySystem.class);
RepositorySystemSession session = MavenRepositorySystemUtils.newSession();

String basedir = singleRepositoryFromListOfDefaultRepositories;
((DefaultRepositorySystemSession)session).setLocalRepositoryManager(system.newLocalRepositoryManager(session, new LocalRepository(basedir)));

ArtifactRequest req = new ArtifactRequest();
req.setArtifact(new DefaultArtifact("commons-io", "commons-io", "jar", "2.5"));

ArtifactResult res = system.resolveArtifact(session, req);

We don't invoke any req.addRepository(repositoryBuilder.build());, so Aether doesn't try to go to any external location.

Line 4 shows that we're trying one of repositories from org.ops4j.pax.url.mvn.defaultRepositories at a time. Each such local repository is checked independently.

org.ops4j.pax.url.mvn.repositories

this is a list of remote repositories searched for an artifact in the second phase of artifact resolution. This repository may contain URIs with file: scheme, but it's better to add such repositories to org.ops4j.pax.url.mvn.defaultRepositories. Each repository is accessed using some configured connector (pax-url-aether uses a connector that underneath invokes httpclient 4.x library - that's why configuring org.apache.http.headers logger may be a good idea)

org.ops4j.pax.url.mvn.localRepository

this is a local repository that supports Aether in second phase of artifact resolution. Its role is a bit different than the role of org.ops4j.pax.url.mvn.defaultRepositories. When Aether actually resolves artifact in one of the remote repositories, it stores the downloaded artifact to org.ops4j.pax.url.mvn.localRepository. That's why write access is required for this location.

If not specified, this property defaults to: ${user.home}/.m2/repository! Be aware of this if Aether seems to find artifacts that are not expected.

Second phase of artifact resolution can be presented using the following code (I love clean code!):

RepositorySystem system = locator.getService(RepositorySystem.class);
RepositorySystemSession session = MavenRepositorySystemUtils.newSession();

String basedir = localRepository;
((DefaultRepositorySystemSession)session).setLocalRepositoryManager(system.newLocalRepositoryManager(session, new LocalRepository(basedir)));

ArtifactRequest req = new ArtifactRequest();
req.addRepository(new RemoteRepository.Builder("ID1", "default", "http://uri1").build());
req.addRepository(new RemoteRepository.Builder("ID2", "default", "http://uri2").build());
req.setArtifact(new DefaultArtifact("commons-io", "commons-io", "jar", "2.5"));

ArtifactResult res = system.resolveArtifact(session, req);

Here, for each remote repository from the list of URIs in org.ops4j.pax.url.mvn.repositories property we call org.eclipse.aether.resolution.ArtifactRequest.addRepository().

Also, in line 4 we use local repository from org.ops4j.pax.url.mvn.localRepository property - the same local repository is used for each remote repository being searched.

org.ops4j.pax.url.mvn.useFallbackRepositories

If true, then Aether will always use http://repo1.maven.org/maven2 repository in addition to any remote repositories specified. I prefer explicit declaration of Maven Central repository (if needed), so it's better to say false here.

org.ops4j.pax.url.mvn.settings

Ah, big topic here. We can specify an explicit location of an XML document following Maven Settings XML Schema.

If not specified, pax-url-aether searches for settings file in the following locations:

${user.home}/.m2/settings.xml
${maven.home}/conf/settings.xml
$M2_HOME/conf/settings.xml

In Karaf 4, implicit location is used (most probably ${user.home}/.m2/settings.xml). In JBoss Fuse, explicit ${karaf.etc}/maven-settings.xml value is configured and default, commented template is shipped.

Why specify custom settings.xml file, when we have properties such as org.ops4j.pax.url.mvn.repositories? There are few things that can be specified only there:

HTTP proxies
custom HTTP headers added when accessing particular remote repositories

Here's the example of HTTP proxy configuration:

<!--
    This is the place to configure http proxies used by Aether.
    If there's no proxy for "https" protocol, proxy for "http" will be used when accessing remote repository
-->
<proxies>
    <proxy>
        <id>proxy</id>
        <host>127.0.0.1</host>
        <port>3128</port>
        <protocol>http</protocol>
        <username></username>
        <password></password>
        <nonProxyHosts>127.0.0.*|*.repository.corp</nonProxyHosts>
    </proxy>
</proxies>

And here's the example of specifying custom HTTP headers:

<!--
    pax-url-aether may use the below configuration to add custom HTTP headers when accessing remote repositories
    with a given identifier
-->
<servers>
    <server>
        <id>maven.central.repo</id>
        <configuration>
            <httpHeaders>
                <httpHeader>
                    <name>User-Agent</name>
                    <value>Karaf</value>
                </httpHeader>
                <httpHeader>
                    <name>Secret-Header</name>
                    <value>secret_value</value>
                </httpHeader>
            </httpHeaders>
        </configuration>
    </server>
</servers>

With custom headers specification, we can see these in logs when accessing repository with ID=maven.central.repo (see below for repository URI specification):

17:30:44,590 | DEBUG | ... | http-outgoing-0 >> GET /maven2/commons-io/commons-io/2.7/commons-io-2.7.jar HTTP/1.1
17:30:44,590 | DEBUG | ... | http-outgoing-0 >> Cache-control: no-cache
17:30:44,590 | DEBUG | ... | http-outgoing-0 >> Cache-store: no-store
17:30:44,590 | DEBUG | ... | http-outgoing-0 >> Pragma: no-cache
17:30:44,591 | DEBUG | ... | http-outgoing-0 >> Expires: 0
17:30:44,591 | DEBUG | ... | http-outgoing-0 >> Accept-Encoding: gzip
17:30:44,591 | DEBUG | ... | http-outgoing-0 >> User-Agent: Karaf
17:30:44,591 | DEBUG | ... | http-outgoing-0 >> Secret-Header: secret_value
17:30:44,591 | DEBUG | ... | http-outgoing-0 >> Host: repo1.maven.org

org.ops4j.pax.url.mvn.repositories - again

After describing org.ops4j.pax.url.mvn.settings let's get back for a moment to org.ops4j.pax.url.mvn.repositories.

If a list of remote repositories in org.ops4j.pax.url.mvn.repositories is prefixed with + sign, all repositories available in all active profiles defined in settings.xml file are appended to effective list of remote repositories searched.

For example if we have this:

org.ops4j.pax.url.mvn.repositories= \
    +http://repo1.maven.org/maven2@id=maven.central.repo

And this in settings.xml:

<!--
    If org.ops4j.pax.url.mvn.repositories property is _prepended_ with '+' sign, repositories from all active
    profiles will be _appended_ to the list of searched remote repositories
-->
<profiles>
    <profile>
        <id>default</id>
        <repositories>
            <repository>
                <id>private.repository</id>
                <url>http://localhost:8181/maven-repository</url>
            </repository>
        </repositories>
    </profile>
</profiles>
<activeProfiles>
    <activeProfile>default</activeProfile>
</activeProfiles>

We can see this in logs during sample resolution:

18:10:17,734 | DEBUG | ... | Using transporter WagonTransporter with priority -1.0 for http://repo1.maven.org/maven2/
18:10:17,736 | DEBUG | ... | Using connector BasicRepositoryConnector with priority 0.0 for http://repo1.maven.org/maven2/
18:10:17,800 | DEBUG | ... | http-outgoing-8 >> GET /maven2/commons-io/commons-io/2.7/commons-io-2.7.jar HTTP/1.1
18:10:17,802 | DEBUG | ... | http-outgoing-8 >> Host: repo1.maven.org
...
18:10:17,872 | DEBUG | ... | Using transporter WagonTransporter with priority -1.0 for http://localhost:8181/maven-repository/
18:10:17,873 | DEBUG | ... | Using connector BasicRepositoryConnector with priority 0.0 for http://localhost:8181/maven-repository/
18:10:17,875 | DEBUG | ... | http-outgoing-9 >> GET /maven-repository/commons-io/commons-io/2.7/commons-io-2.7.jar HTTP/1.1
18:10:17,876 | DEBUG | ... | http-outgoing-9 >> Host: localhost:8181
...

org.ops4j.pax.url.mvn.globalChecksumPolicy

When Aether fetches artifact from remote repository, it always tries to download SHA1/MD5 checksum for the artifact. It may fail to do so. If repository URI doesn't specify per-repository value, this global property's value is used. Actually if this global value is specified, per-repository values are ignored.

This property may have 3 values determining Aether's behavior:

fail - resolution fails
warn - information is printed at WARN level
ignore - nothing happens.

Note that there's no way to prevent fetching checksums.

org.ops4j.pax.url.mvn.globalUpdatePolicy

When Aether fetches SNAPSHOT artifacts, it needs to fetch maven-metadata.xml first. Before hitting org.ops4j.pax.url.mvn.repositories, Aether checks the presence of resolver-status.properties file in org.ops4j.pax.url.mvn.localRepository location (this status file is specific to given groupId, artifactId and version, for example: <REPOSITORY>/commons-io/commons-io/2.5-SNAPSHOT/resolver-status.properties). We can control whether Aether actually should refresh metadata information:

always - Aether always fetches maven-metadata.xml when resolving SNAPSHOTs
never - opposite of the above
daily - Aether fetches maven-metadata.xml if a day passed since timestamp written in maven-metadata-ID_OF_REPOSITORY.xml.lastUpdated property inside resolver-status.properties file.
interval:<NUMBER_OF_MINUTES> - Aether fetches maven-metadata.xml if given number of minutes passed.

Maven Repository URI

I've mentioned repository URI in few places. When specifying URI on a org.ops4j.pax.url.mvn.repositories list, we may use the following format:

http(s)://host:port/path@snapshots@noreleases@id=ID@other_options

Options that may be specified are:

id=ID - this option may (should) be specified to identify a repository. We may then refer to the repository for example when specifying custom headers.
snapshots - whether the repository should be used when resolving SNAPSHOT artifacts
noreleases - whether the repository should not be used when resolving non-SNAPSHOT artifacts
releasesUpdate=daily|never|always|interval:MINUTES - see description of org.ops4j.pax.url.mvn.globalUpdatePolicy property
snapshotsUpdate=daily|never|always|interval:MINUTES - see description of org.ops4j.pax.url.mvn.globalUpdatePolicy property
update=daily|never|always|interval:MINUTES - see description of org.ops4j.pax.url.mvn.globalUpdatePolicy property
releasesChecksum=fail|warn|ignore - see description of org.ops4j.pax.url.mvn.globalChecksumPolicy property
snapshotsChecksum=fail|warn|ignore - see description of org.ops4j.pax.url.mvn.globalChecksumPolicy property
checksum=fail|warn|ignore - see description of org.ops4j.pax.url.mvn.globalChecksumPolicy property

Other options

There are other properties that can be configured in org.ops4j.pax.url.mvn PID:

org.ops4j.pax.url.mvn.defaultLocalRepoAsRemote: Whether local repository specified in org.ops4j.pax.url.mvn.localRepository should be added as first remote repository inserted to the list configured with org.ops4j.pax.url.mvn.repositories property - it's a bad idea...

Caveats

Due to highly asynchronous nature of OSGi™ (and in particular - a slight race condition between pax-url-aether that configures org.ops4j.pax.url.mvn.MavenResolver service on one side and felix.fileinstall and felix.configadmin bundles that create configuration for org.ops4j.pax.url.mvn PID on other side), there's short period of time where other configuration may be used in org.ops4j.pax.url.mvn.MavenResolver service.

To prevent such interregnum, I suggest duplicating Maven/Aether properties from ${karaf.etc}/org.ops4j.pax.url.mvn.cfg in ${karaf.etc}/config.properties. If pax-url-aether can't find ConfigurationAdmin (yet), it defaults to bundle properties and these may be specified in etc/config.properties.

Summary

I hope the above information will clear all confusion related to pax-url-aether configuration in OSGi framework as JBoss Fuse or Karaf.

Programming, contributing

2016-07-22

Using Maven with OSGi Part 2