go/extra/conbench: Add consensus benchmark command

[abukosek]

No description provided.

[codecov]

Codecov Report

Merging #3183 into master will decrease coverage by 0.22%.
The diff coverage is 0.00%.

@@            Coverage Diff             @@
##           master    #3183      +/-   ##
==========================================
- Coverage   66.02%   65.80%   -0.23%     
==========================================
  Files         371      371              
  Lines       33285    33287       +2     
==========================================
- Hits        21978    21906      -72     
- Misses       8074     8134      +60     
- Partials     3233     3247      +14     

Impacted Files	Coverage Δ
go/oasis-node/cmd/common/pprof/pprof.go	19.64% <0.00%> (-0.73%)	⬇️
go/worker/compute/executor/committee/batch.go	66.66% <0.00%> (-16.67%)	⬇️
go/worker/compute/executor/committee/state.go	71.42% <0.00%> (-14.29%)	⬇️
go/consensus/tendermint/abci/state/state.go	54.54% <0.00%> (-9.10%)	⬇️
go/consensus/tendermint/epochtime/epochtime.go	85.18% <0.00%> (-7.41%)	⬇️
go/worker/common/committee/runtime_host.go	65.04% <0.00%> (-4.86%)	⬇️
go/worker/compute/executor/committee/node.go	62.19% <0.00%> (-3.08%)	⬇️
go/storage/mkvs/lookup.go	72.63% <0.00%> (-2.11%)	⬇️
go/consensus/tendermint/abci/mux.go	56.74% <0.00%> (-2.06%)	⬇️
go/storage/client/client.go	69.85% <0.00%> (-1.92%)	⬇️
... and 21 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update bb1aa48...b487942. Read the comment docs.

[pro-wh]

so far so good

[pro-wh]

a little reminiscent of https://github.com/oasisprotocol/oasis-core/blob/master/go/oasis-node/cmd/debug/txsource/workload/parallel.go the parallel workload. this is more robust to failures and adds measurements so we can get a transaction result out. a long time ago we were using a scheme where we'd take the middle 80% of the transactions to get the peak rate.

oasis-core/clients/utils/src/benchmark.rs

Line 105 in dce901f

// Cut 10% at the beginning and the end.

[abukosek]

Thanks for taking a look at this. I can add the transaction cutting, but IMO it's not needed -- in my test runs, only the first block at the start of the benchmark had fewer transactions, the other blocks were always full. As long as there are enough samples (so that the duration of the benchmark is long enough), it won't significantly affect the results.