7.0 Proposal: BasicPublish with ROM<byte> for exchange and routing key

[bollhals]

The basic publish is one of the "hottest" method we have, so any improvements there are surely very valuable.

Extracted from #970, I propose to extend the IModel interface with a new method for 7.0:
void BasicPublish(ReadOnlyMemory<byte> exchange, ReadOnlyMemory<byte> routingKey, bool mandatory, IBasicProperties basicProperties, ReadOnlyMemory<byte> body);

This replaces the expensive string -> byte encoding for the often times same string.

[stebet]

I'm all for performance, but I don't really see this as something that will be widely used, but for some very niche cases. IO will always be the bottleneck, not CPU usage.

[stebet]

I've just had a second thought about this, and I've gone 180 on my original comment. I think this might be a clever idea for high-perf scenarios. I also remembered that there exists an API that does similar things: https://github.com/StackExchange/StackExchange.Redis/blob/main/src/StackExchange.Redis/Interfaces/IDatabase.cs#L241

The StackExchange.Redis API uses RedisKey structs as key parameters which have implicit conversion to both byte[] and string: https://github.com/StackExchange/StackExchange.Redis/blob/main/src/StackExchange.Redis/RedisKey.cs#L178-L191

That means that clients that use hardcoded strings for exchanges/queues etc. can define a ready-made ReadOnlyMemory<byte> instance that can reused without the repeated string -> byte[] cost.

[bollhals]

To indicate the caching intent for this, we could also define the signature as
void BasicPublish(RabbitMqString exchange, RabbitMqString routingKey, bool mandatory, IBasicProperties basicProperties, ReadOnlyMemory<byte> body);
where RabbitMqString is defined as

public readonly struct RabbitMqString 
{ 
    public readonly string Value;
    internal readonly ReadOnlyMemory<byte> CachedBytes;
    public RabbitMqString(string value) { /* implement */ };
}
/**************** OR *****************/
public sealed class RabbitMqString 
{ 
    private ReadOnlyMemory<byte> _cache;
    public string Value { get;}
    internal ReadOnlyMemory<byte> CachedBytes => /* populate _cache if not empty or return _cache */;
    public RabbitMqString(string value) { /* implement */ };
}

where the struct would build the cache within the ctor, while the class could defer it to it's first use if we want to.

[bollhals]

Update from a local running version on this:

Method	Mean	Error	StdDev	Gen 0	Gen 1	Gen 2	Allocated	Code Size
BasicPublishWrite	41.3951 ns	0.8661 ns	0.0475 ns	-	-	-	-	386 B
BasicPublishSize	18.5197 ns	0.1749 ns	0.0096 ns	-	-	-	-	178 B
BasicPublishMemoryWrite	14.3104 ns	0.6103 ns	0.0334 ns	-	-	-	-	775 B
BasicPublishMemorySize	0.8075 ns	0.0310 ns	0.0017 ns	-	-	-	-	39 B

Obviously the gain increases the larger the string (Exchange / RoutingKey) is. (Benchmark is with "Exchange_OR_RoutingKey").

Another benchmark to serialize into the 3 frames of a message:

Method	Mean	Error	StdDev	Ratio	Gen 0	Gen 1	Gen 2	Allocated	Code Size
BasicPublish	223.9 ns	13.50 ns	0.74 ns	1.00	-	-	-	-	2504 B
BasicPublishMemory	182.1 ns	7.83 ns	0.43 ns	0.81	-	-	-	-	2518 B

It my does not seem huge, but it increases the throughput on the most relevant hotpath.
I'll setup a PR for 7.0 in the next days.