In a clustered environment when calling Liferay services, requests can take up to 15-30 seconds due to cache replication

• Type: Bug
• Status: Closed
• Resolution: Duplicate
• Affects Version/s: 7.0.X EE, Master
• Fix Version/s: None
• Component/s: Performance
• Labels: None
• Fix Priority: 4

Reproduction steps:

1. Make sure DXP has at least 2G for heap
2. Set up DXP to be clustered (one node is sufficient, as issue happens on cache sending side) (e.g. set cluster.link.enabled=true)
3. Start DXP on de-14 or de-20
4. Deploy the attached jars
5. Log in with test/test
6. Add the deployed plugins to the page and click on "Add" (just to compile the jsps)
7. Open the attached TEST_SB_v1.1_oneloginonly.jmx script with JMeter (note: increase JMeter memory to 1-2Gb)
8. If "Logout" is enabled, disable it
9. Start the script

Experienced behavior: Take a couple of thread dumps and check the occurrences of the "CoalescedPipe.put" string. After a while, it should round up to 70 - number of threads executing by JMeter. In the reports on JMeter, you'll see that the timing of the "Add entry" step will go into tens of seconds (average, max, etc).

Additional information:
1. Reproduced on de-14 and de-20 and de-21
2. LPS-69744 is very similar, but that's already included in de-11.
3. Reproducible with both unicast and multicast

Cannot test on ee-7.0.x: Threads get blocked waiting for MySQL to respond, "Lock Timeout Exceeded" from MySQL in log start appearing after about a minute. If test is stopped, portal returns to normal after 10 seconds or so.