What to consider to tune modcluster configuration in wildfly10

[wwang2016]

Hi

I was able to configure wildfly 10 modcluster to work with Apache mod_cluster (1.3.1). However, when I was doing a load test, I found out that the test through web server eventually caused error in wildfly instance and I also saw error log in Apache web server

The obvious error in wildfly instance is the so-called "java.net.SocketException: Too many files open". When I use the command lsop -u   | grep TCP | wc -l, I can see the number grew steadily until the wildfly instance reported the error. This was when I sent requests through web server.

However, when I sent the requests through wildfly instance (app server) directly, the number did not grow, and the app server can take a lot heavier load without this issue.

I guess some configuration on modcluster (wildfly) needs to be tuned? What are some common recommendations on tune modcluster on wildfly configuration (standalone-full-ha.xml)?

Thanks

Wayne

[mbabacek]

HI Wayne, this looks like a nasty resource leak, not a matter for tuning. May I ask you to create a mod_cluster - JBoss Issue Tracker  JIRA issue with configuration files, logs and the OS platform, please? You could assign to me and I'll investigate.

Thank you for your time

Cheers

-K-

[wwang2016]

Hi Michal,

This issue seemed a bit complicated.

This issue can be replicated after the Apache server is up and running (no restart) and I do many rounds of load tests. For example, If I start up the web server, and then I do many rounds of load test today. Then I will easily replicate this issue with just one round of load test (even with smaller load) the second day.

However, if restart the Apache web server, and do a fresh load test, this issue does not happen.

I will try a more consistent approach to replicate this issue, and prepare the configuration files, log files, and then I will create the issue

Thanks,

Wayne

[wwang2016]

Hi Michal,

I created the issue under mod_cluster. The issue is MODCLUSTER-536. It got assigned automatically to Jean-Frederic Clere, and I could not change the assignee. You may just re-assign it to yourself.

I did many rounds of load test yesterday, and the typical JMeter setting was 50 users x 2000 loops with ramp-up = 0. It was pretty demanding load. I did around 10 rounds of tests.

Please let me know if you need any other information.

Thanks,

Wayne

[mbabacek]

THX Wayne, it's in the pipeline now. At a glance, the intel on the Jira looks sufficient.

Cheers

-K-

[wwang2016]

Hi Michal,

I just did four rounds of load tests which were a bit less demanding than yesterday

50 users x 2000 loops with ramp-up = 1 seconds, and I also duplicated this issue.

Thanks,

Wayne