Reduce EventListener overheads

[MihaZupan]

• When writing events with no payload (just event id), avoid allocating an empty object[] and ReadOnlyCollection for the EventWrittenEventArgs payload
• In WriteEventWithRelatedActivityIdCore, avoid fetching metadata from m_eventData[eventId] multiple times
• In DecodeObject, remove the const bool m_EventSourcePreventRecursion = false and the logic around flipping a m_EventSourceInDecodeObject ThreadStatic we would always ignore because of m_EventSourcePreventRecursion anyway
• Avoid fetching payload type from within DecodeObject itself, pass it in instead (so we don't re-fetch metadata for each parameter)
• Instead of recalculatnig the expected parameter count on every call to WriteToAllListeners, do it once and store the value in EventMetadata (EventListenerParameterCount)

Net result: save 2 object allocations when writing events with no payload, save some cycles in all cases

Benchmark.cs

Method	Toolchain	Mean	Error	StdDev	Median	Ratio	Gen 0	Allocated
WriteEventNoParams	base	79.86 ns	1.524 ns	1.426 ns	79.50 ns	1.00	0.0440	184 B
WriteEventNoParams	new	64.50 ns	1.227 ns	2.844 ns	64.01 ns	0.78	0.0324	136 B
WriteEventIntParams	base	198.66 ns	3.903 ns	6.837 ns	195.68 ns	1.00	0.0668	280 B
WriteEventIntParams	new	134.20 ns	2.711 ns	3.619 ns	133.30 ns	0.67	0.0668	280 B
WriteEventStringParams	base	527.01 ns	6.637 ns	6.208 ns	526.14 ns	1.00	0.0744	312 B
WriteEventStringParams	new	433.64 ns	5.706 ns	5.338 ns	433.57 ns	0.82	0.0744	312 B

[ghost]

Tagging subscribers to this area: @tarekgh, @tommcdon, @pjanotti
See info in area-owners.md if you want to be subscribed.

Issue Details

---

• When writing events with no payload (just event id), avoid allocating an empty object[] and ReadOnlyCollection for the EventWrittenEventArgs payload
• In WriteEventWithRelatedActivityIdCore, avoid fetching metadata from m_eventData[eventId] multiple times
• In DecodeObject, remove the const bool m_EventSourcePreventRecursion = false and the logic around flipping a m_EventSourceInDecodeObject ThreadStatic we would always ignore because of m_EventSourcePreventRecursion anyway
• Avoid fetching payload type from within DecodeObject itself, pass it in instead (so we don't re-fetch metadata for each parameter)
• Instead of recalculatnig the expected parameter count on every call to WriteToAllListeners, do it once and store the value in EventMetadata (EventListenerParameterCount)

Net result: save 2 object allocations when writing events with no payload, save some cycles in all cases

Method	Toolchain	Mean	Error	StdDev	Median	Ratio	Gen 0	Allocated
WriteEventNoParams	base	79.86 ns	1.524 ns	1.426 ns	79.50 ns	1.00	0.0440	184 B
WriteEventNoParams	new	64.50 ns	1.227 ns	2.844 ns	64.01 ns	0.78	0.0324	136 B
WriteEventIntParams	base	198.66 ns	3.903 ns	6.837 ns	195.68 ns	1.00	0.0668	280 B
WriteEventIntParams	new	134.20 ns	2.711 ns	3.619 ns	133.30 ns	0.67	0.0668	280 B
WriteEventStringParams	base	527.01 ns	6.637 ns	6.208 ns	526.14 ns	1.00	0.0744	312 B
WriteEventStringParams	new	433.64 ns	5.706 ns	5.338 ns	433.57 ns	0.82	0.0744	312 B

Author:	MihaZupan
Assignees:	-
Labels:	area-System.Diagnostics.Tracing
Milestone:	6.0.0

[noahfalk]

Thanks @MihaZupan, looks good to me aside from tweaking the comment

[noahfalk]

cc @josalem @davmason @sywhang

[noahfalk]

Thanks for the perf improvements @MihaZupan!
cc @stephentoub