[ 
https://issues.apache.org/jira/browse/MRESOLVER-278?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17617958#comment-17617958
 ] 
Hudson commented on MRESOLVER-278:
----------------------------------

Build succeeded in Jenkins: Maven » Maven TLP » maven-resolver » master #58

See 
https://ci-maven.apache.org/job/Maven/job/maven-box/job/maven-resolver/job/master/58/

> BREAKING: Make Session extend AutoCloseable (and introduce onCloseHandlers)
> ---------------------------------------------------------------------------
>
>                 Key: MRESOLVER-278
>                 URL: https://issues.apache.org/jira/browse/MRESOLVER-278
>             Project: Maven Resolver
>          Issue Type: New Feature
>          Components: Resolver
>            Reporter: Tamas Cservenak
>            Assignee: Tamas Cservenak
>            Priority: Major
>             Fix For: 1.9.0
>
>
> So far, in (vanilla) Maven, the lifecycle of session was on par with 
> lifecycle of SISU container, as Maven does something like this:
>  * boot, create container
>  * create session
>  * work
>  * destroy container
>  * exit JVM
> So, Maven execution is 1 session 1 container, are on par.
> This is not true for cases where container (and resolver components) are 
> reused across several sessions, like mvnd does. Also, current code on master 
> (named locks adapter) uses {{@PreDestroy}} to shut down adapters, that is 
> invoked when container is shut down, while the adapters are created 
> per-session. This means that long-living mvnd daemons will shut down the 
> unused adapter only when daemon itself is shut down, even is session for 
> which adapter was created is long gone/done.
> While Maven has "session scoped" notion, resolver has not. Hence, simplest 
> and cleanest solution is to make RepositorySystemSession have a method to 
> "close", denoting that "this session is done". Also, if we can provide hooks 
> for "onSessionClose", this resolves all the problems, as for example the 
> adapter, that is created per session, could be now cleanly shut down at 
> session end.
> One gotcha: this change implies a {*}breaking change for integrators of 
> resolver{*}:  integrator should make sure they close the session after they 
> are done with it.
> Example changes needed for resolver users: 
> [https://github.com/apache/maven/pull/822]
> The "pattern" to make use of this feature in resolver is following:
>  * stuff something into session data, ideally using computeWhenAbsent
>  * if absent, register a callback hook as well
>  * in callback, get the stuffed thing from session and if present, do 
> something with it



--
This message was sent by Atlassian Jira
(v8.20.10#820010)

Reply via email to