RFR: 8335771: Improve stability of java/nio/channels/DatagramChannel tests

[Daniel Fuchs]

On Mon, 8 Jul 2024 11:36:01 GMT, Alan Bateman <alanb at openjdk.org> wrote:

>> Several `java/nio/channels/DatagramChannel` tests fail intermittently. Though it seems difficult to fix all intermittent failures due to the nature of what some of these attempt to test, the stability can still be improved.
>> 
>> The most common source of intermittent failures are:
>> 
>>  - not receiving a datagram due to port conflict or interference of concurrent running tests
>>  - failure to disconnect due to interference of concurrent running tests
>>  - failure due to reception of unexpected stray datagrams
>> 
>> Avoiding to bind on the wildcard address, filtering unexpected datagram, and some retrying can aleviate some of the issues and greatly reduce the frequency of intermittent failures.
>
> test/jdk/java/nio/channels/DatagramChannel/NotBound.java line 29:
> 
>> 27:  *     the channel's socket to be bound to a local address. This test may fail
>> 28:  *     intermittently on macOS if other datagram channel tests are running
>> 29:  *     concurrently on the same host.
> 
> I don't think we should add this comment. Instead, if there is interference possible on the same host then we should create a TEST.properties in this directory with `exclusiveAccess.dirs=.` so that the tests don't run concurrently.

OK. Note that interference can also come from other tests in other directories (like DatagramSocket tests in jdk/java/net/..) or from system processes (like syslogd on macOS which binds to an ephemeral port).

> test/jdk/java/nio/channels/DatagramChannel/SelectWhenRefused.java line 65:
> 
>> 63:             if (((InetSocketAddress)dc.getLocalAddress()).getPort() == port) {
>> 64:                 if (i < MAX_TRIES - 1) {
>> 65:                     System.out.format("Refuser port has been reused by dc: %s, retrying...%n",
> 
> I think this test is a JUnit test, in which case trace messages should probably go to System.err so that they are inlined with other test output rather than in a separate section of the .jtr file.

Good point.

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/20050#discussion_r1668820050
PR Review Comment: https://git.openjdk.org/jdk/pull/20050#discussion_r1668820419