Observed on vm215-m3, Dart's "dartium-win-ia32-stable" bot
- https://uberchromegw.corp.google.com/i/client.dart/builders/dartium-win-ia32-be
- https://uberchromegw.corp.google.com/i/client.dart/builders/dartium-win-ia32-stable

A disk usage alert for drive "C:\" fired. Closer inspection revealed ~38G (of 60G C drive) consumed with "scoped_dir" directories under "c:\Users\chrome-bot\AppData\Local\Temp" several hours old (read: older than the current test suite). Contents are:

scoped_dir10420_18107/:
Cache  databases

scoped_dir10420_18107/Cache:
data_0  data_1  data_2  data_3  index

scoped_dir10420_18107/databases:
Databases.db  Databases.db-journal

These directories:
a) need to be cleaned up at the beginning of new test runs,
b) should not be leaked at the end of current test runs, and
c) ideally should not exist on the C drive.

(a) and (b) are self-explanatory, so quickly on (c): on most infrastructure bots, each system has a small OS disk (C) and a large test disk. Data written to the test disk is generally far more resilient to disk full issues and can be managed easier via periodic pruning scripts. Data written to C runs the risk of overwhelming the small disk.

If possible (understood that it's not always possible b/c of OS configuration), these "scoped_dir..." should be written to a subdirectory on the large testing disk.

I'm not sure where the problem here lies, so CC sheriffs + some Dart team.