-
Type: New Feature
-
Resolution: Fixed
-
Priority: Major - P3
-
Affects Version/s: None
-
Component/s: Internal Code
-
None
-
Fully Compatible
-
v4.0
-
Service Arch 2019-03-25
-
8
Currently our fast clock is implemented by a background thread which wakes up every 10ms if callers continue to check it's time. If that background thread get's descheduled for a long time for some reason, this can cause the background threads time to drift substantially.
If we:
- force every read of date_t::now to update a new date_t::lastNow
- change the background thread so that it's responsibility is to wake up and call Date_t::now() (if readers are still present)
- change the background thread clock source now() to return date_t::lastNow
We'll get a world in which we have the same target granularity + thread quiescence as before, but additionally ensure that even if the background thread can't be scheduled, that lastNow() will still be pushed loosely forward if there are callers of Date_t::now() elsewhere in the system. This should ensure a tighter bound on the spread between the precise and fast clock sources
- related to
-
SERVER-40165 generate wtimeout deadlines with the precise clock
- Closed