Optimize history calculation to reuse previously computed results

[hannesg]

The history algorithm currently recalculates the whole history for the repository on every run. However, most of the things in a repo do not change on each commit. I introduced a cache that memorizes the last ten calculation results so that we can reuse them later. This bring the history calculation in most cases down from O(N) to O(1) for N = number of commits in repo.

Benchmark:

/build/services/cd-service/pkg/history $ go test -benchmem -bench=.
goos: linux
goarch: amd64
pkg: github.com/freiheit-com/kuberpult/services/cd-service/pkg/history
cpu: Intel(R) Core(TM) i7-10750H CPU @ 2.60GHz
BenchmarkHistoryNoCache-12             4         289058522 ns/op        36413094 B/op    2330716 allocs/op
BenchmarkHistoryCache-12             465           2560525 ns/op          596726 B/op      23879 allocs/op
PASS
ok      github.com/freiheit-com/kuberpult/services/cd-service/pkg/history       45.740s

Those who do not --learn-from-- cache history are doomed to repeat it.

Lenin

[medo-freiheit]

I like the idea and the new LRU part looks great, but I got confused with history structure, so maybe @sven-urbanski-freiheit-com could review this