Open Stack

On 2014-06-05 19:50:30 -0700 (-0700), Kevin Benton wrote:
[...]
> At a queue size of 20 and an 80% success rate. The patch in
> position 20 only has a ~44% chance of getting merged. However,
> with a queue size of 4, the patch in position 4 has a ~71% chance
> of getting merged.
[...]

Your theory misses an important detail. A change is not ejected from
a dependent Zuul queue like that of the gate pipeline simply for
failing a voting job. It only gets ejected IF all the changes ahead
of it on which it's being tested also pass all their jobs (or if
it's at the head of the queue). This makes the length of the queue
irrelevant to the likelihood of a change eventually making it
through on its own, and only a factor on the quantity of resources
and time we spend testing it.

The reason we implemented dynamic queue windowing was to help
conserve the donated resources we use at times when there are a lot
of changes being gated and failure rates climb (without this measure
we were entering something akin to virtual memory swap paging
hysteresis patterns, but with virtual machine quotas in providers
instead of virtual memory).
-- 
Jeremy Stanley