Roman Grytskiv
Roman Grytskiv
Just hit same issue.
@deniszh Thanks for the feedback! > I'm afraid hash will be constantly changing and coordinator will be forced to "fix" all existing files that way or another, Not sure if...
@deniszh > Both will help with NULLs but not with wrong aggregation. Absolutely. See model points 3 and 4. Thing is, if we can repair data before roll-ups, they will...
@deniszh > correction process can be long running process and loosing all progress because of restart could be not really nice. Again, in case of Cassandra repair e.g. [Cassandra Reaper](http://cassandra-reaper.io/)...
@flucrezia > Maybe a different approach could be a randomised algorithm; in terms of scanning time, past time window and metrics to compare. Randomisation should help in efficiency without loosing...
@flucrezia > I agree with you as long as the assumption that we need to check all the data holds. But I don't think that is the case because a...
This will have to wait until we have a more-or-less clean lint run. Then we can do it right away.
@jwkohnen We definitely know that context cancellations fail from time-to-time. We can clearly see the cases when the requests do not return after the request timeout is exceeded (which causes...
Here's an illustration of what happens https://play.golang.org/p/V4riLAtcUAq
@ksurent Thanks for the suggestions. We will be able to compare after we do the refactoring. I think, in this case, we need something simple. Just a classic limited-length requests...