-
Bug
-
Resolution: Unresolved
-
Minor
-
OS: Amazon Linux 2
Jenkins version: 2.223
Java version: 8.212.04-r0
JAVA_OPTS: "-Djava.awt.headless=true -XX:ActiveProcessorCount=48 -Dhudson.slaves.NodeProvisioner.initialDelay=0 -Dhudson.slaves.NodeProvisioner.MARGIN=50 -Dhudson.slaves.NodeProvisioner.MARGIN0=0.85"
Installed plugins: in comment
How Jenkins is accessed: Through an AWS Classic Load Balancer
Jenkins installation method: Docker image jenkins/jenkins:2.223-alpine running on ECS, agents launched on-demand by ECS Plugin and connecting via JNLP
Web browser: Various users affected, all using modern versions of Chrome/Firefox
OS: Amazon Linux 2 Jenkins version: 2.223 Java version: 8.212.04-r0 JAVA_OPTS: "-Djava.awt.headless=true -XX:ActiveProcessorCount=48 -Dhudson.slaves.NodeProvisioner.initialDelay=0 -Dhudson.slaves.NodeProvisioner.MARGIN=50 -Dhudson.slaves.NodeProvisioner.MARGIN0=0.85" Installed plugins: in comment How Jenkins is accessed: Through an AWS Classic Load Balancer Jenkins installation method: Docker image jenkins/jenkins:2.223-alpine running on ECS, agents launched on-demand by ECS Plugin and connecting via JNLP Web browser: Various users affected, all using modern versions of Chrome/Firefox
About once a week, multiple threads handling HTTP requests matching the patterns /blue/rest/organizations/jenkins/pipelines/<folder>/pipelines/<job>/runs/<number>/ or /blue/organizations/jenkins/<folder>%2F<job>/detail/<job>/<number>/pipeline/ will get stuck in an infinite loop in StandardGraphLookupView.bruteForceScanForEnd.
Each thread that get stuck ties a CPU core, and they tend to happen in batches of about 20-30 threads. This results in the Jenkins web UI appearing to be very slow to users. The resolution so far has been to use the JavaMelody monitoring plugin to kill the stuck threads one by one.
I've attached stack traces from two example threads showing that they get stuck in the same place, as well as a snapshot of the CPU usage at the time. There was nothing supicious going on at the time, and I've been unable to manually reproduce the issue.
I believe something like https://github.com/francoiscampbell/workflow-api-plugin/compare/master...graph-lookup-concurrent should address the issue, but I'm not sure how to verify that.
