[WIP] Remove ToArray calls on MemoryStream

[Mpdreamz]

and try to reuse .Buffer, new serializer can handle any padding at the end.

[Mpdreamz]

Working on a continuation of this PR where I update our RecyclableMemoryStream with recent changes from https://github.com/microsoft/Microsoft.IO.RecyclableMemoryStream which actually implements TryGetBuffer() rather then falling back to its base MemoryStream implementation.

Something in the implementation of TryGetBuffer() seems off, either it needs to account for offset or the length is not correct if we are reusing blocks.

Some more investigation is needed

[russcam]

Is it safe to call TryGetBuffer on a RecyclableMemoryStream? It looks like this method is not overloaded.

[Mpdreamz]

It's safe in the sense that it works but see my comment on the PR, updating to a later version that actually implements it yields some problems under heavy usage.

I have a feeling it sometimes writes at an offset which trygetbuffer does not take into account, not been able to trap it yesterday though

[russcam]

Is this still waiting on some changes, @Mpdreamz?

[Zyklop]

@Mpdreamz In a performance analysis, I found this to be a problem in our use-case. Do you need some help, or testing assistance? I think it could make a noticeable difference in performance.

[Mpdreamz]

Hi @Zyklop we are meeting as a team next week in person and spending some time doing some performance deep dives.

I've opened #4191 and #4172 as continuations of this branch.

It does not tackle everything we want to investigate but its a first step.

If you like to share your analysis results feel free to email me them to ml at elastic.co, more data/visuals to look at the better.

Any testing you can do is helpful and appreciated as always, keep an eye out next week.

[Mpdreamz]

Closing this as its out of date.