Hide Forgot
Date of First Response: 2007-11-20 08:39:02 project_key: SOA Problem description: This bug blocks us from running ESB stress tests for longer than 20 minutes. The bug was originally reported in JBESB-880. If between 10 and 100 messages per second generated, the ESB server becomes unresponsive and the consoles are locked. If the traffic stream is stopped, the message queue may be process and the server may recover, but sometimes it hangs indefinitely. This condition - a large queue backlog - can happen if an ESB app has an outage. Or, is a client app sends messages faster then ESB reads them from queue - for example, a batch process - such as a job that starts up at midnight and generates a large number of messages Test description: The test exercises clear throughput of the ESB - no influence of alien code - the server deployed is a no-op/placeholder. The test: 1) Generates big amount of messages to the JMS GW queue 2) Deploys the ESB application 3) When the first message is received the internal message counter and timer is started 4) When the last message is processed the speed is calculated and test is ended Steps to run test: 1) Obtain the ESB bits 2) Obtain dbinstall (see below) 3) Setup and run dbinstall to have custom ESB config for database 4) Obtain quickstart-based tests from https://svn.corp.jboss.com/repos/soa/trunk/qa/tests/ 5) Setup build.properties in tests/quickstart/tests and point it on ESB bits and created db configuration 6) Run ant rebuild-ng - it will build framework, tests and setup ESB 7) Run the test with 10 messages 8) If the test is successful, rerun with larger numbers of messages You can also check the script that is used for hudson matrix build project jboss-soa-platform-IntegrationTest-testsuite-matrix - it performs the same setup steps) Test source: https://svn.corp.jboss.com/repos/soa/trunk/qa/tests/quickstarts/performance1 Documentation: https://svn.corp.jboss.com/repos/soa/trunk/qa/tests/quickstarts/tests/performance_tests.txt DB setup scripts: https://svn.corp.jboss.com/repos/soa/trunk/qa/dbinstall/ - just edit the build.properties and run ant (Test developed by Jirka Pechanec - thanks Jirka!)
Link: Added: This issue depends JBESB-880
Link: Added: This issue is related to JBQA-1029
Link: Added: This issue is a dependency of SOA-84
Link: Removed: This issue is a dependency of SOA-84
Link: Added: This issue related SOA-84
Link: Added: This issue related SOA-93
Link: Removed: This issue related SOA-84
Kevin (paraphrasing) "on going, work progressing and checking some other performance issues."
fixes merged in from ESB head - revision 17282, if performance needs revisiting please reopen against beta1
I run Helloworld performance test in optimized sending mode over night. The server was able to process 1,000,000 messages within 31546843ms = 8.76 hours. This makes average throughput 31.7 messages per second (unfortunately I left file log level to DEBUG and ended with >1GB log file, I have to rerun it). After that the server was perfectly responsive, no errors appeared on the output and the server was able to gracefully terminate.
The speed with file log level set to WARN is definitely better (~50 msgs/s).
Verified.
Link: Added: This issue related SOA-877