100% CPU usage on 3.10.1

Hi there!

I’ve just applied the update from 3.10.0. to 3.10.1.

I’ve downloaded the deb file, stopped openfire and installed using “dpkg -i openfire_3.10.1_all.deb”.

First thing i noticed: nothing is complaining about java nice work here!

No “–ignore-depends=openjdk-7-jre” was needed.

But shortly after the update is saw this:

And from what I can tell there is still a Bug with 100% CPU usage.

I did have a quick look with strace on the crazy thread:

read(217, “”, 128)

= 0

gettimeofday({1434534355, 485756}, NULL) = 0

gettimeofday({1434534355, 485798}, NULL) = 0

gettimeofday({1434534355, 485815}, NULL) = 0

epoll_wait(49, {{EPOLLIN|EPOLLERR|EPOLLHUP, {u32=217, u64=16704004399789244633}}}, 4096, 1000) = 1

this repeats very fast. maybe its a indication for whats going on.

Let me know how i can help you with logs or stack traces.

Kind regards

I upgrade from 3.10.0 to 3.10.1 with rpm package.

But, I’m sorry to say failed.

This time I got “High CPU usage and hangup” in 30 mins.

Should I downgrade to 3.9.3?

To clarify, your users can not log in when this happens or is just excessive CPU usage being reported?

Users can still log on, just like on 3.10.0.

On the Openfire Admin Console, when you go to the Sessions Tab, do you see a session with an invalid hostname stated for the “Client IP”?

nope.

right now there are only three connections, but all of them have a valid ip.

Tracking this issue here [OF-920] 100% Usage with thread attempting to read from a should be closed session - Jive Software Open Source

1 Like

Update:

Downgrade didn’t work, update to 3.10.1 again.

I found that users can logon and search who is connected via ssl port 5223 ,when users connected via 5222(tls) failed.

So I disabled tls at 5222, restart openfire service. All users can logon and search now.

running openfire on debian wheezy x64 with oracle java 8 (userd to be 7)

Less than 20 users, usually there are ~5 people online. most are using android client xabber.

auth via LDAP, using mysql for DB

active plugins: monitoring for chat archive and kraken for irc/icq transport (only 2 people are using transports)

no BOSH or S2S connections, one MUC room.

had 3.9.3 running for almost a year, never had trouble.

when 3.10.0 was released I upgraded. Since the upgrade openfire rapidly gets to 100% cpu usage. Usually 2-3 days after staring the service it was so severe that the monitoring was raising alarms, openfire needed to be restarted.

yesterday I upgraded to 3.10.1. less than 24hrs later openfire is again causing 100% cpu load.

Now, I dont want to say it got worse but it sure diddnt get better.

Unfortunately I dont have any knowledge about debugging or java and have a hard time understanding the logs either so I seem not able to give much more information.

I just downgraded back to 3.9.3 because restarting openfire every few hours is a less favourable solution for me than running 3.9.3

But please, I would like to have a 3.10.x running stable and without eating up all my cpu power in the future. this problem seems to me to be somewhat common, so I have hope that we can get enough information that it can and will be fixed soon.

thanks.

Catwiesel

Fixing this is very high priority, our current thought is to downgrade the bundled MINA to version 2.0.7 and Tom has prepared a pull request to do just that.

OF-922: Revert MINA to 2.0.7 by tevans · Pull Request #241 · igniterealtime/Openfire · GitHub

We just need some testing of this patch to see if it breaks anything else in the process.

1 Like

thats good to hear. unfortunately I cant really help in the process but I sure do appreciate the work of all the devs and testers.

fingers crossed and looking forward to a functioning openfire 3.10.x

catwiesel

Okay, we have a build with MINA reverted to 2.0.7, could you folks please try this and report back your findings!

Debian

http://bamboo.igniterealtime.org/artifact/OPENFIRE-DB3/shared/build-7/Project-De bian-distribution-files/openfire_3.10.2_…

RPM

http://bamboo.igniterealtime.org/artifact/OPENFIRE-RB3/JOB1/build-11/Project-RPM -installer-files/openfire-3.10.2.alpha-1…

Windows

http://bamboo.igniterealtime.org/artifact/OPENFIRE-WB3/shared/build-8/Project-Wi ndows-distribution-files/openfire_3_10_2…

daryl

2 Likes

installed 3.10.2.alpha

monitoring system load. expect my results to be posted in ~ 36hrs
sooner if problems arise

Hi there!

Thanks for the quick work! I’ve just installed the 3.10.2 alpha. Let’s have a look

3.10.2 seems to do the trick. so far no excessive cpu load.
will continue to monitor and report back if something should happen.

Hi there!

Version Openfire 3.10.2 Alpha is running smooth since 48 hrs.

Thanks for your quick help!

3.10.2 is now released. I am going to lock this thread, so please reopen new threads if you happen to see this issue again!