On Tue, 17 May 2022 12:45:52 GMT, Daniel Jeliński <djelin...@openjdk.org> wrote:
> This PR improves the execution time of jdk_net tests (and, by extension, > tier2) by about 3 minutes. > > Tests located under `jdk/java/net/httpclient/websocket` are never run in > parallel. Each of the 8 modified `Pending***` tests originally required 40 > seconds to complete. After the proposed changes, they usually complete in 15 > seconds. > > This PR modifies the tests to initially run with 1 second timeout. If the > test fails with 1 second timeout, it is retried with timeout increased to 10 > seconds (the original value). > > The modified tests were executed at least 10 times on each of: Windows, Linux > (both x64 and aarch64), MacOS (both x64 and aarch64). No failures were > observed. I have poor recollection of the mechanics of these tests. The source of wait is timed get and `assertHang`. You are reducing the former. That looks good. What looks questionable is rearrangement of asserts: when `assertHangs` moves down. assertNotDone(cfClose) can transitorry pass even if ping has not hung. Tiriviality: since you are here, please remove 1. superfluous "are" on BlowupOutputQueue:60 2. "the a" on BlowupOutputQueue:63 test/jdk/java/net/httpclient/websocket/PendingOperations.java line 44: > 42: static final Class<IOException> IOE = IOException.class; > 43: // Time after which we deem that the local send buffer and remote > 44: // receive buffer must be full. This has been heuristically > determined. "Heuristically" was a weird word to use here; should've been "empirically", I think. You deleted the whole sentence, which is okay too. ------------- Changes requested by prappo (Reviewer). PR: https://git.openjdk.java.net/jdk/pull/8746
