Daniel, please run some numbers on what would happen if we process both halves of all hang pairs. I'm particularly interested in: - capacity implications for HBase - statistical bias introduced into aggregate reports on Socorro
The problem is that when we do that, the calculation of a total crash rate becomes messed up (or at least significantly more complicated), as then some part of the release reports would be throttled and some wouldn't be. Also, hangs would probably always bubble up to the top of topcrasher lists on releases, as they'd have a factor 10 more reports as crashes with the same rate.
KaiRo: in the discussion that lead to this bug filing, we were unable to come up with an algorithm that didn't introduce some sort of bias. The point of this bug is to figure out the impact of the bias to see if we can compensate for it.
Lars, sure, I was was just showing up two cases that cause problems here and that need to be looked at in the research being done for comment #0.
Closing out old issues. Not in current scope of work.