The fix for https://issues.jenkins-ci.org/browse/JENKINS-63014 introduced a regression bug. This fix was released in Jenkins version 2.248 and broke a feature used in e.g. yet-another-docker plugin.
Our Jenkins master is exposed via multiple URLs:
- primary URL: this is the publicly used URL for the Jenkins server, our users and some of the slave agents only capable of reaching Jenkins master through this address
- secondary URL: this is an internal URL (only reachable from the host machine that runs the Jenkins master), some of the slave agents are only capable of reaching Jenkins master through this address
We're using yet-another-docker plugin with JNLP launcher to launch Docker slaves. It has an option called "Different Jenkins master URL". This option can be used to overwrite the primary URL (the one set in Jenkins global config). Since version 2.248 Jenkins doesn't respect this secondary URL and the slaves fail to connect to Jenkins master.
Log from previous versions:
Log from 2.248 version: