Closed kchodorow closed 12 years ago
Cool. When we add filters that handle unique events like those, I'll go back and add them into the clock skew module.
clock_skew.py now uses the new algorithm that we discussed yesterday. It logs each clock skew value it finds, along with a corresponding weight for that value, in a dictionary and stores that information in a collectionName.clock_skew collection of the db. Also, if the algorithm finds a clock skew value that matches a majority of the entries, it adds that value and stops looking for more possibilities.
Right now, it is still failing one test that it shouldn't be failing, but otherwise seems to be working as expected.
Note: clock_skew.py should also use exit messages to help determine clock skew values, probably with a larger margin for error to accomodate slower shutdowns. Right now, it does not do this.
...also kidding
not kidding!!
None of the filters do it yet, but some filters will identify unique events across servers that will make clock skew detection easier (e.g., issue #24, replica set reconfigs: each reconfig has a unique id, so we know that reconfigs took place at the same time, at least among up servers). I think movechunk will do this too, probably some others.