Serguey Asael Shinder: The JVM stops printing stack traces on repeated exceptions
A note from a few days ago, kept deliberately: by the time you read this it should have dropped out of news-sitemap.xml, because Google only wants URLs from the last two days in there. The page itself stays online and stays in the ordinary sitemap. Only the news metadata ages out.
The substance, briefly. The JVM omits stack traces for hot, repeatedly-thrown exceptions — -XX:-OmitStackTraceInFastThrow turns that off. If you have ever seen a NullPointerException in production with no frames at all, that is what happened: the first few had traces, the rest were optimised away.
Two consequences worth holding on to.
- The trace you are missing was probably in the first occurrence. Search your logs backwards, not forwards.
- Do not leave the flag off in production to be safe. It exists because building traces for a tight loop of exceptions is expensive. Turn it off while you diagnose, then turn it back on.
Part of the running notes: Written for the person debugging at three in the morning