wombatu-kun opened a new pull request, #16786: URL: https://github.com/apache/iceberg/pull/16786
ADLSInputStream and ADLSOutputStream capture a full thread stack trace (Thread.currentThread().getStackTrace()) in their constructors. The stack is only used by finalize() to report the creation site when a stream is garbage collected without being closed, but the capture runs on every stream creation, i.e. on the per-file read/write path. ResolvingFileIO already records the creation stack of the FileIO it delegates to, and tells the delegate to skip its own capture by setting init-creation-stacktrace=false in the delegate's properties. ADLSFileIO did not honor that flag, so every ADLS stream re-captured a stack trace even though ResolvingFileIO had already recorded one. This makes the ADLS streams honor init-creation-stacktrace (default true), read through AzureProperties, exactly as #16739 does for the S3 input and output streams. The problem and the fix are identical to #16739; that PR contains the JMH benchmarks quantifying the per-stream allocation saved by skipping the capture. When the capture is disabled, finalize() still logs an unclosed-stream warning and points to the property to re-enable the creation stack. Tests: TestAzureProperties covers the default, the explicit-disable value that ResolvingFileIO relies on, and that the setting survives serialization. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
