There are several cases in WiredTiger where a timestamp is used by multiple threads. Since time is not co-ordinated amongst threads (we don't guarantee that it can't go backward), we need to review those usages and ensure they are safe.
Ideally we'll add a mechanism for making it unlikely we'll introduce bugs in shared-time functionality in the future.
Original description:
sue.loverso notes:
I'm sure there are threads that can be negatively affected by time moving backward in that they'll compute a wildly long wait time.
Based on the discussion in WiredTiger Pull Request #3537, we should guarantee atomic update of clock times before threads act on them.