Now that the linux-386-longtest builder is correctly running the tests in 386 mode (as of CL 234520), it seems to be shaking out some flakiness in TestSmhasherWindowed (CC @randall77):
2020-06-01T20:38:23-13bc6d4/linux-386-longtest
2020-05-29T20:56:20-6ad5f4e/linux-386-longtest