Workspaces stuck consuming 100% CPU since 5.17

[davidwindell]

We are experimenting with v5.17.0. It seems a bug has been introduced that causes che to get stuck (indefinitely) consuming 100% of the CPU when retrieving the project file tree.

I can reliably reproduce this by switching between v5.16.0 (where it doesn't happen) to v5.17.0 where it does.

The process in question is the main workspace process:

/usr/lib/jvm/java-1.8.0-openjdk-amd64/bin/java -Dnop -Djava.util.logging.manager=org.apache.juli.ClassLoaderLogManager -Xms256m -Xmx2048m -Djava.security.egd=file:/dev/./urandom -Dche.logs.dir=/home/user/che/ws-agent/logs -Dche.logs.level=INFO -Djuli-logback.configurationFile=file:/home/user/che/ws-agent/conf/tomcat-logger.xml -Djdk.tls.ephemeralDHKeySize=2048 -Djava.protocol.handler.pkgs=org.apache.catalina.webresources -agentlib:jdwp=transport=dt_socket,address=4403,server=y,suspend=n -Dcom.sun.management.jmxremote -Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.authenticate=false -Dche.local.conf.dir=/mnt/che/conf -classpath /home/user/che/ws-agent/conf/:/usr/lib/jvm/java-1.8.0-openjdk-amd64/lib/tools.jar:/home/user/che/ws-agent/bin/bootstrap.jar:/home/user/che/ws-agent/bin/tomcat-juli.jar -Dcatalina.base=/home/user/che/ws-agent -Dcatalina.home=/home/user/che/ws-agent -Djava.io.tmpdir=/home/user/che/ws-agent/temp org.apache.catalina.startup.Bootstrap start

I'm happy to provide access to che devs to the workspace in question if that helps debug the issue.

Large project trees seem to be the most affected.

[ghost]

@davidwindell to reproduce it I need to start Che v 5.17, import a big project (btw, can you suggest one?) and expand it in the project tree?

[davidwindell]

Here's a screenshot from Che v5.17 (note the XHR request takes 1 minute just to get the project tree).

And here's a screenshot from Che v5.16 (exactly the same project, folder, network, etc)

Now only a few ms.

[ghost]

@davidwindell thanks. I can confirm that the request to project api in 5.16 and 5.16 is completed differently...5 seconds vs milliseconds.

[ghost]

Btw, deleting git repo fixes the problem, but that's, of course, not a solution.

[vparfonov]

Possibly related to #5722
@vinokurig can you check?

[vinokurig]

analyzing