While investigating IMPALA-5889, I was able to reproducibly crash Impala on latest master. The crash doesn't repro on 2.9 so I suspect it's a regression. I will add reproduction steps shortly.
Segmentation fault in ScannerContext::Stream::GetBytesInternal(long, unsigned char**, bool, long*)
Scanners should include file and offset information in errors
Add truncating files to test_scanners_fuzz.py
Remove defensive nullptr check in ScannerContext::Stream::GetNextBuffer()
Make all errors coming out of the IoMgr identifiable
Avro scanner can crash when HDFS seek fails