[
https://issues.jboss.org/browse/WFWIP-276?page=com.atlassian.jira.plugin....
]
Jean Francois Denise commented on WFWIP-276:
--------------------------------------------
A possible way to solve this performance issue would be to not start CLI JVM and embedded
server. The running server would handle the configuration at boot time. This needs a deep
analysis, but mainly the startup sequence would become:
* /opt/eap/bin/openshift-launch.sh
* Launch scripts generate CLI script in well known location (already the case).
* Server starts, a hook (TBD) applies the CLI script (require CLI operation to ModelNode
parsing).
Boot time of application built via S2I is much longer than in
before-Galleon times
----------------------------------------------------------------------------------
Key: WFWIP-276
URL:
https://issues.jboss.org/browse/WFWIP-276
Project: WildFly WIP
Issue Type: Bug
Components: OpenShift
Reporter: Jan Blizňák
Assignee: Jean Francois Denise
Priority: Critical
Attachments: test-app.zip
As a consequence of Galleon RFEs EAP7-891 and mainly EAP7-1216 there is big difference in
pod with EAP application start up time which requires S2I configuration.
This is an overview of quick comparison of average boot times of the application pods of
the attached test app (we are not interested in build/deploy pod times etc):
||17.0-6||pod time||EAP boot||config time
|#1 | 8283|6128|2155
|#2 | 9446|7548|1898
|#3 | 8767|6871| 1896
|#4 | 8568|6272| 2296
|#5 | 8000 |6362|1638
||average ||8613ms||6636ms||1977ms
||18.0-7||pod time||EAP boot||config time
|#1 |13769|5716|8053
|#2 |13362|5952|7410
|#3 |14261|6093|8168
|#4 |14322|6135|8187
|#5 |13921|6326|7595
||average||13927ms||6044ms||7883ms
pod time = time from the pod start until EAP is fully booted (log contains {{started in
..}})
EAP boot = time of boot EAP itself == {{started in ..}}
config time = the difference of previou timess = anything before EAP boot begins
Again we are talking here about the time needed for prepared application image to full
start. We can clearly see the EAP boot times are comparable between 17 and 18 images but
the configs times are almost 4 times longer with CD18 images.
This is very unfortunate as in my point of view the pod boot time is the most critical
time for our users - they would typically prepare the app images only once in a time but
their application can scale up and down many times during its uptime.
The cause of this that when S2I is needed with image 18.0+ the part of the configuration
that was previously done during S2I build just once is now processed on *each* pod start.
--
This message was sent by Atlassian Jira
(v7.13.8#713008)