[BUG] EventHub's retry method always adds 4 seconds to a configured delay

[adriannowak]

Describe the bug EventHub SDK fails to deliver retry messages at configured rate. Seems it just ignores the delay which is set in AmqpRetryOptions.getDelay() and by some reason adds extra time SERVER_BUSY_WAIT_TIME. This property is not configurable, and is always 4 seconds. In a consequence SDK delivers messages slower than it is expected.

Exception or Stack Trace Add the exception log and stack trace if available N/A

To Reproduce Steps to reproduce the behavior:

The following unit test should pass after at most 5 seconds (1 failure + 2 retries + timeout 3x500). With the latest version of SDK following code takes 16 seconds.

    @Test
    @Timeout(value = 5)
    void withRetryMono() {
        // Arrange
        final String timeoutMessage = "Operation timed out.";
        final AmqpRetryOptions options = new AmqpRetryOptions()
            .setMaxRetries(2)
            .setTryTimeout(Duration.ofMillis(500))
            .setMode(AmqpRetryMode.FIXED);

        final AtomicInteger resubscribe = new AtomicInteger();
        final Mono<AmqpTransportType> neverFlux = TestPublisher.<AmqpTransportType>create().mono()
            .doOnSubscribe(s -> resubscribe.incrementAndGet());

        StepVerifier.create(RetryUtil.withRetry(neverFlux, options, timeoutMessage))
            .expectSubscription()
            .expectErrorSatisfies(error -> assertTrue(error.getCause() instanceof TimeoutException))
            .verify();

        assertEquals(options.getMaxRetries() + 1, resubscribe.get());
    }

Code Snippet Add the code snippet that causes the issue.

This breaking change was introduced in this PR

https://github.com/Azure/azure-sdk-for-java/blob/main/sdk/core/azure-core-amqp/src/main/java/com/azure/core/amqp/implementation/RetryUtil.java#L109

        final Duration delay = options.getDelay().plus(SERVER_BUSY_WAIT_TIME);

Expected behavior A clear and concise description of what you expected to happen.

SDK retries the delivery at a configured pace.

        final Duration delay = options.getDelay();

Information Checklist Kindly make sure that you have added all the following information above and checkoff the required fields otherwise we will treat the issuer as an incomplete report

• [x] Bug Description Added
• [x] Repro Steps Added
• [x] Setup information Added

[joshfree]

Thanks for filing this issue, @adriannowak. @conniey could you take a look at this regression?

/cc @Azure/azsdk-sb-java

[adriannowak]

Hey @joshfree @conniey - Could you please give any update on the issue? When will be the fix available?

[conniey]

Thanks for reporting this. The SERVER_BUSY_WAIT_TIME should only be applied in cases where the exception is a "ServerBusyException" to align with behaviour in the legacy library.

That PR was part of a retry policy cleanup in 2021.

[anuchandy]

Note: given the environment is experiencing server-busy throttling from the EH service, the overall slowness may not change much since the SDK (after any fix) is still has to use back-off on server-busy.

The SDK requirement to apply back-off on server-busy is not a regression, this is a recommended pattern, which exists in all generations of Event Hubs library (including track1 3.4.x).