PgPoolTest.testEventLoopSize fails on AMD 9950X CPU

[adamdickmeiss]

Version

5.0.5

Context

Test PgPoolTest.testEventLoopSize fails on AMD 9950X CPU.

Problem: in testEventLoopSize , then num becomes 64 (2 * CpuCoreSensor.availableProcessors()) Reducing this value to 50 makes the test pass. This is probably due to default connection limit in Postgres of 100 = 2 * 50.

Steps to reproduce

mvn compile
cd vertx-pg-client
mvn test

Produces:

*** Iteration 49/50 of test testNoConnectionLeaks(io.vertx.tests.pgclient.PgPoolTest)
*** Iteration 50/50 of test testNoConnectionLeaks(io.vertx.tests.pgclient.PgPoolTest)
[ERROR] Tests run: 28, Failures: 0, Errors: 1, Skipped: 0, Time elapsed: 24.06 s <<< FAILURE! -- in io.vertx.tests.pgclient.PgPoolTest
[ERROR] io.vertx.tests.pgclient.PgPoolTest.testEventLoopSize -- Time elapsed: 0.105 s <<< ERROR!
io.vertx.pgclient.PgException: FATAL: sorry, too many clients already (53300)

[INFO] 
[INFO] Results:
[INFO] 
[ERROR] Errors: 
[ERROR]   PgPoolTest.testEventLoopSize » Pg FATAL: sorry, too many clients already (53300)
[INFO] 
[ERROR] Tests run: 28, Failures: 0, Errors: 1, Skipped: 0

[tsegismont]

Does it fail consistently? Is this the only error message?

Regardless the value of num, this test only creates two connections. Perhaps another test doesn't close connections properly...

[adamdickmeiss]

It fails consistently and if I run it in isolation with mvn -Dtest=PgPoolTest#testEventLoopSize test it fails also.

Perhaps try to reproduce by forcing num=50 in a local test (if fewer CPU cores on system)?

The error message most likely really is Postgresql server reporting with the "unique" 53300 ID.

[tsegismont]

Right, my bad, I misread the test. It uses at most 2 event loops but the number of connections is num*2.

[adamdickmeiss]

thx @tsegismont for dealing with this.

[tsegismont]

Thank you for taking time to report it

[tsegismont]

Fixed by #1572