Hi Charlie, Thanks for finding and fixing this! I have fixed many such failures in the logging tests in the past.
Your changes look good to me. Reviewed. Do you need a sponsor for this? best regards, -- daniel On 22/07/2020 20:38, Charlie Gracie wrote:
Hi, I believe there was a subtle timing hole introduced into the HttpURLConWithProxy.java test with the fix for JDK-8183369 [1]. In our local tests, when using GC stress modes, we consistently get a failure reported as: "Execution failed: `main' threw exception: java.lang.RuntimeException: Connection not retried with proxy". I verified that the Proxy is being used but the messages are not being logged with the ProxyHandler. The issue is caused by back to back calls of Logger.getLogger(String). If a GC happens between the two calls there is no guarantee that the same object will be returned [2]. Since the test relies on the log messages being handled by the new ProxyHandler to pass, it needs to ensure the Logger object lives the entire lifetime of the test. Here is my proposed fix: http://cr.openjdk.java.net/~burban/cgracie/logger_fix/webrev0.0/ The easiest way to reproduce is to use the following vm options when running the specific test. These options will force the Shenandoah GC into a continuous GC mode making the test fail almost 100% of the time. -vmoption:-XX:+UseShenandoahGC -vmoption:-XX:+UnlockDiagnosticVMOptions -vmoption:-XX:ShenandoahGCHeuristics=aggressive Thanks, Charlie Gracie [1] https://bugs.openjdk.java.net/browse/JDK-8183369 [2] https://docs.oracle.com/en/java/javase/14/docs/api/java.logging/java/util/logging/Logger.html#getLogger(java.lang.String)
