So you should not be checking that the slave is launched by running through computer.connect(false) as that requires that you return early. Instead you should be checking for e.g. the ssh service to be open and responding to connection requests (you can do that easily with your launcher... but a custom check would be better as you only want to check that the port is open) then you return and let Jenkins add the node and connect. The critical bug is https://github.com/jenkinsci/jclouds-plugin/blob/master/jclouds-plugin/src/main/java/jenkins/plugins/jclouds/compute/JCloudsCloud.java#L297 i.e. you should *never* add the node to jenkins yourself, when you do that you cause the second issue that you are then trying to correct.
In short the future should return the Node instance to be added *when* the actual Node instance is ready to connect. If you return the Node too soon, then you get over-provisioning... which then leads to people trying to "fix" that by copying the hacks in the EC2 plugin. The cloud API is a mess. I want to replace it with something that a sane individual has a hope of implementing correctly. Do not fear, after dealing with the cloud API for many years now I have come to the conclusion that it is IMPOSSIBLE to write a correct implementation. The closest thing to a correct implementation is the oc-cloud implementation used by CJOC (and that only avoids the bugs that you cannot avoid because we can linearize operations in CJOC). Our nectar-vmware plugin is "ok" (largely due to a shake-down we did when we were using it for the predecessor to CJOC) The next closest to a correct implementation is Docker. After that... well it's all mostly sh1t... I loose the will to live when I start looking at other implementations. If you pick EC2 as your template you will be following every wrong example in the book. On 27 April 2016 at 13:54, Oliver Gondža <ogon...@gmail.com> wrote: > On 04/27/2016 02:08 PM, Stephen Connolly wrote: > >> >> On 8 April 2016 at 14:44, Oliver Gondža <ogon...@gmail.com >> <mailto:ogon...@gmail.com>> wrote: >> >> Some plugins even wait for launch to complete before leaving >> PlannedNode#future >> >> >> Those are broken plugins >> > > As a maintainer of one such plugin I would like to hear your input on [1] > is it seems to be quite common. With regards to the cloud-stats plugin, it > would not change much even if they all get fixed since launch can be > started by Jenkins before it collects completed futures, it just makes it > happen all the time instead of rarely. > > [1] > https://github.com/jenkinsci/jclouds-plugin/blob/master/jclouds-plugin/src/main/java/jenkins/plugins/jclouds/compute/JCloudsCloud.java#L299-L303 > > -- > oliver > > -- > You received this message because you are subscribed to the Google Groups > "Jenkins Developers" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to jenkinsci-dev+unsubscr...@googlegroups.com. > To view this discussion on the web visit > https://groups.google.com/d/msgid/jenkinsci-dev/418b92f5-e0ca-c85b-ecf2-bc4d5a184422%40gmail.com > . > > For more options, visit https://groups.google.com/d/optout. > -- You received this message because you are subscribed to the Google Groups "Jenkins Developers" group. To unsubscribe from this group and stop receiving emails from it, send an email to jenkinsci-dev+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/jenkinsci-dev/CA%2BnPnMzkwFR60CEnd8XuVwcYOJaH0PRfSavRmy_31bnqgq0qqQ%40mail.gmail.com. For more options, visit https://groups.google.com/d/optout.
