Vladislav

1. We should not print tons of logs in case of cancellation

Last two comments are still open

> Why not put them into the shared place and the last flow will execute them? The issue I'm thinking about is how to provide backpressure on on other flows,...

> Special flow for global command: Basically, how it works now. But we can't get the same problem because it always locks the full db and out of order execution...

@uhthomas Hi. We just newly support this behavior, but don't have yet an explanation for it. Dragonfly by default disallows accessing undeclared keys in Lua scripts. You can enable it...

You can run memtier on a master instance. Because it has back pressure it will slow down if the replica does not catch up. The issue is that the bottleneck...

Because when I ran the benchmarks: - Skipping execution at all did not increase throughput - Improving async streamer wakeups increased throughput a lot (so there is opportunity for improvement...

Yes, I didn't mean the bandwidth directly, just the whole mechanism that is responsible for sending replication data

Is this the time it takes the benchmark to run on master? Its strange there is such a large difference. It can be caused by the replica buffer filling up....

if you run `INFO ALL` some sections are titled IN CAPS others are Normal