search

WASdev / ci.docker.websphere-traditional

Dockerfiles for WebSphere Application Server traditional
Apache License 2.0
170 stars 190 forks source link

slow startup of tWAS container in ICP #176

Closed donbourne closed 4 years ago

donbourne commented 5 years ago

First launch of traditional WAS container after deployment takes about 15 minutes to reach "ready for e-business". "Scale up images" take roughly 4 minutes.

THEORY: seems to be 5 java processes running at same time during startup, and only the logViewer limits its memory usage with -Xmx. So it may be a memory problem related to change in default setting for java 8 to use -XX:+UseContainerSupport by default (see https://www-01.ibm.com/support/docview.wss?uid=swg21672834#java8sr5fp26)

Some of the java processes running concurrently at startup:

helpful to run this from inside the container during startup: watch 'ps -C java -o pid,cmd,rss,time'

jwalcorn commented 4 years ago

I found I had to increase the memory limit for the deployment (in the yaml) a lot, compared to what I use for Liberty. That's probably stating the obvious, but just in case - upping that like from 256MB to 1GB make my tWAS container start a LOT faster (in OpenShift, in my case)

gmarcy commented 4 years ago

that seems comparable to our experience with the helm chart, was-liberty uses:

resources: constraints: enabled: false requests: cpu: 500m memory: 512Mi limits: cpu: 4000m memory: 2Gi

and was-traditional uses:

resources: constraints: enabled: false requests: cpu: 500m memory: 2Gi limits: cpu: 4000m memory: 10Gi

or a 4x to 5x increase in the memory limit

gmarcy commented 4 years ago

Closing issue since the current helm charts increased the memory limits for tWAS.