Details
-
Bug
-
Status: Resolved
-
Critical
-
Resolution: Cannot Reproduce
-
Private Beta
-
None
-
None
Description
I ran perf record on a TS running the "real world TPCH" benchmark, and saw a weird stack trace with CFileIterator::Scan -> memcpy(). This function doesn't really do any memcpy as far as I knew, so I digged a bit deeper. It looks like it's spending a lot of time memcpy()ing a Status object.
Seems like ultra low-hanging fruit for scan perf.