Erik Grinaker

>Can you try running BenchmarkDistSenderSunnyDay as well. Only did 3 runs since the benchmarks were pretty slow, so used `-delta-test none`. Overhead is about as expected, except for the high-count...

@nvanbenschoten I added the context timeout check we discussed, such that we won't set up a cancellable context if the context already has a timeout below the probe threshold +...

Verified that this recovers `failover/non-system/deadlock/lease=expiration` in less than 10 seconds.

Isn't this check in place to ensure that followers' logs don't regress? Do we have any other safeguards for that? We don't check that during `MsgApp` because the follower's log...

Yep, you're right, nice find! ``` toydb> create table a (id int primary key, value string not null); Created table a toydb> create table b (id int primary key, value...

This should be addressed now, but I'm going to have a closer look at this logic and add more test cases in a while.

Thanks!

@ahrtr Hey, sorry for the review lag here, it's been a crazy week. I'll be on PTO for the next week, but maybe @pav-kv can take a look.

I share @pav-kv's concerns in https://github.com/etcd-io/raft/pull/169#issuecomment-1983994248, and also the concerns voiced in #167. Spurious election disruption is a known weakness when not using pre-vote and checkquorum, and it sounds like...

How does this interact with CheckQuorum+PreVote? Leader stability can be important for availability under certain kinds of network partitions -- it looks like this will still respect the leader recency...