Flaky test fixed

[Derrick2000]

Issue: https://issues.redhat.com/browse/UNDERTOW-2528

Dear Maintainer,

I noticed that one of the WebSocket tests in the repository is exhibiting flaky behavior due to Non-Order-Determinism (NOD) issues. Below is a detailed explanation of the problem and a proposed fix to ensure the test behaves reliably.

Occasionally, the test fails at the line:

undertow/websockets-jsr/src/test/java/io/undertow/websockets/jsr/test/reconnect/ClientEndpointReconnectTestCase.java

Line 117 in b44a1c3

Assert.assertEquals("MESSAGE-ECHO-hi", endpoint.message());

with the error:
expected:<[MESSAGE-ECHO-hi]> but was:<[CLOSE]>

The issue stems from Non-Order-Determinism (NOD) in the asynchronous WebSocket communication. Specifically:

1. Race Condition During Reconnection: After sending the "close" message, the test immediately proceeds to send "hi" without ensuring that the WebSocket connection has been fully reopened.
   The "CLOSE" message from the previous interaction might still be processed or pending when "hi" is sent, causing the test to retrieve "CLOSE" instead of "MESSAGE-ECHO-hi".

2. Session State Assumption: The test assumes that the Session is immediately ready to send and receive messages after receiving an "OPEN" message, which may not be the case in all scenarios.

3. Unreliable Message Processing Order: WebSocket communication is asynchronous, and the test does not synchronize properly with the server's message processing, leading to unexpected order of messages.

Proposed Fix
To resolve the issue, I propose the following change: Verify session state before sending messages by adding a check to ensure the Session is open and ready for communication. This change ensure the test behaves reliably by addressing the Non-Order-Determinism issue caused by race conditions in the asynchronous WebSocket communication.

Please let me know if you’d like further clarification or assistance with implementing these changes.

Best regards,

[baranowb]

Hi @Derrick2000 . Ive created ticket for this enchancement: https://issues.redhat.com/browse/UNDERTOW-2528 . Since you already analyzed the problem, would you like to contribute said fix?

[Derrick2000]

Hi @baranowb, our Nondex tool detected flakiness in the test io.undertow.websockets.jsr.test.reconnect.ClientEndpointReconnectTestCase.testAnnotatedClientEndpoint. The fix I implemented in this PR addresses and resolves this issue.