Strange yearly recurring problem with performance


#1

This strange thing that keeps occurring yearly around this time of year (around the end of March, Easter, switch to summertime). Then we suddenly get major performance issues with our actively used Lucee application. Nothing particular in scheduled tasks, higher traffic etc., it just starts and after a few days (around 2 days) it goes away. For the last years our solution was to add a lot more CPU / hardware during that time, but that’s not a real solution. It’s happening right now again. Do you have any ideas that might lead us to a solution.


#2

did restarting the server make a difference? also which version of lucee?


#3

I’m afraid we’ve had to do quite a bit of restarting because of the problem. Because all the users are then logged out, the problem is temporarily away, but it starts building up as users start logging on again. So a restart does not provide a solution. We are currently on Lucee 5.1.4.19.


#4

are you able to test 5.2.6.60? or does it only occur under load?


#5

Yes, only under load. It is however a good idea to migrate to a higher version. But nonetheless, the problem remains that this occurs every year around the same date, whatever the version.


#6

Just a me too with this issue, with version 4.5. We have two servers with scheduled tasks and both began to increase cpu 24h before the summertime change.

We have identify that the problem was with high context switching:

imagen

Restart of lucee, even restart of the server doesn’t solve the problem.

In fusionReactor we can see that there are blocked thread :

The solution for us is remove the schuduler tasks and create again.


#7

Very interesting! I’m not sure I understand what high context switching means, but I’ll look into it. I understand it helped you to recreate the scheduled tasks. Do you mean you deleted every one of them and then recreated every one of them, or was it enough to restart / pause the scheduled tasks? (the reason I ask is because we have quite a lot of scheduled tasks).


#8

In a linux box you can see the context switching with the command: sar w 2 20

We need to remove and create again the tasks. We also have a lot of them, but we create it with cfschedule:

https://cfdocs.org/cfschedule


#9

what version of java are you all running?


#10

Hi Zac,

openjdk-8-jre 8u151 in ubuntu 16.04


#11

Thanks David, I’ll look into context switching. The tip for using cfschedule is also great. With so many tasks, it’s hard to keep the oversight. Cfschedule will definitely help with that!


#12

1.8.0.161 SE Runtime Environment 64-bit Server VM mixed mode.


#13

it has everything to do with how JAVA handles time change. Reset your server or java services affected by the time change and the memory usage will drop.


#14

I rebooted the server without success. Only remove/create the schudler tasks works for us.


#15

time to file a bug in jira?


#16

Yes, I also think that it has to do with the time change. We had the same problem last year and this year again on one of our servers. The performance went up to 100% and only reboot of the machine solved the problem.