I've run the impala minicluster with TSAN occasionally when trying to debug races. I think it would be useful to run the backend tests with it. The other annoyance with TSAN is that there are a number of false positives with the JVM that we'd need to suppress.
It seems worth the effort. It would also be useful to finish off enabling leak sanitizer for the handful of backend tests that don't have it enabled. I turned it on for the vast majority of backend tests but there were a few that still leaked thrift servers and things like that. On Mon, May 6, 2019 at 11:13 AM Todd Lipcon <t...@cloudera.com> wrote: > That's great to hear. Has anyone tried running any BE tests with TSAN in > recent years? It's been invaluable for Kudu, and while I'm sure it would > take a bit of effort to get passing on Impala, the ongoing returns would be > worth it IMO. > > One point, though, is that for TSAN to work well, you need to instrument > thirdparty libs as well, so we'd need to start from native-toolchain and > work our way up. > > -Todd > > On Mon, May 6, 2019 at 10:49 AM Tim Armstrong <tarmstr...@cloudera.com> > wrote: > > > Hi All, > > I turned on backend tests under UBSAN in our precommit tests - Jim > Apple > > let me know that they were passing after he fixed the final set of issues > > there. > > > > - Tim > > > > > -- > Todd Lipcon > Software Engineer, Cloudera >
