vitorenesduarte / fantoch Goto Github PK

-C debug-assertions can be passed to the compiler to have debug_assert! executed in optimized builds.

Hello,

Great work here!

I am interesting in creating a simple distributed KV store using Tempo, but I ran into some issues when testing. First I don't really understand some flags:

Why are these required? and what is the purpose?
faults, processes

What is the purpose of the these?
sorted_processes, NFR

Also It would be nice, if you could document how to run it. For example, a bash script that runs all nodes locally.

Thanks!

Is there any documentation/tutorial on how to run the simulator locally?

Your dependency file specified a branch or reference for https://github.com/vitorenesduarte/dashmap, but Dependabot couldn't find it at the project's source. Has it been removed?

View the update logs.

https://github.com/stateright/stateright
https://github.com/rystsov/fast-jepsen

• mit-pdos/noria@42aa4cb
• https://github.com/gnzlbg/jemallocator/blob/master/jemalloc-sys/README.md#cargo-features

Try compiling with -C opt-level=3 -C target-cpu=native -C codegen-units=1 (as suggested here) and/or RUSTFLAGS="--emit=asm" (as suggested here) and check differences.

• https://github.com/tikv/raft-rs
• https://github.com/zowens/paxos-rs

https://github.com/jonhoo/griddle

Currently, if some transient error occurs during an experiment, the full set of experiments will hang forever. It should be possible to set a MAX_EXPERIMENT_TIME after which the experiment is restarted (or maybe skipped after MAX_EXPERIMENT_RETRIES).

• Should also improve the cleanup before starting the experiment:

  • kill any previous process/client
  • kill dstat
  • and so on

• Could also show a "progress bar" with:

  • "3 of 7" done
  • "average experiment time: 1m30s"
  • "expected to end in: 6m"

@vitorenesduarte This is an insanely great piece of work, especially if Tempo turns out to be bug-free.
(This is outside my area, so, I can't evaluate that. )

One thing that would be fantastic to see in addition to the six protocols implemented, is the popular 'Raft'.
Maybe I am missing a reason it wouldn't fit into this evaluation framework, but it seems this would be
enhanced a lot if a Raft back-end could be implemented.

We should at least support a form of batching that does not affect latency: the next time the coordinator wakes up, it should take all new issued commands and form a batch with them. Generalized protocols may be affected by this since a batch can contain conflicting commands.

• check https://github.com/metrics-rs/metrics
• make metrics visible to tests and refactor tests in https://github.com/vitorenesduarte/fantoch/blob/master/fantoch_ps/src/protocol/mod.rs

Your dependency file specified a branch or reference for https://github.com/vitorenesduarte/dashmap, but Dependabot couldn't find it at the project's source. Has it been removed?

View the update logs.