Summary data: Notes based on observation of telemetry data & code

[vladsud]

1. I was looking at summarize telemetry code, and I'm not sure I follow this logic:

        const summarizeEvent = PerformanceEvent.start(logger, {
            eventName: "Summarize",
            refreshLatestAck,
            ...summarizeTelemetryProps,
        });

            const generateSummaryEvent = PerformanceEvent.start(logger, {
                eventName: "Summarize",
                ...summarizeTelemetryProps,
            });

Why do we need to events with exactly same name in same function? I think you can leverage first one just fine - they represent the same starting point in time.

2. I'm not sure there is a value in recording constants:

message: "summaryAck",

3. Why Whiteboard data is different from Chapter1? opsSinceLastSummary & timeSinceLastAttempt is missing:
   union Office_Fluid_FluidRuntime_*
   | where Event_Time > ago(7d)
   | where Data_eventName == "fluid:telemetry:Summarizer:Running:Summarize_end"
   | extend WB = Data_hostScenarioName contains "Whiteboard"
   | summarize count(), avg(Data_timeSinceLastAttempt), avg(Data_opsSinceLastSummary), avg(Data_totalBlobSize), avg(Data_duration)
   by Data_eventName, WB

Data_eventName	WB	count_	avg_Data_timeSinceLastAttempt	avg_Data_opsSinceLastSummary	avg_Data_totalBlobSize	avg_Data_duration
fluid:telemetry:Summarizer:Running:Summarize_end	false	1,580,649	8450170	30	117863	2518
fluid:telemetry:Summarizer:Running:Summarize_end	true	388,603	21678480	NaN	NaN	3533

Note that if we look at fluid:telemetry:Summarizer:Running:GenerateSummary data, we will see that Whiteboard has both opsSinceLastSummary & timeSinceLastAttempt payload. So maybe it's a mix of different versions and their format that screws the data.

4. I'm not sure how to interpret timeSinceLastAttempt - see above: average is in millions!

[NicholasCouri]

@vladsud

#1. Done.

#2. there is no need to have that constant summaryAck

#3 As for avg_Data_opsSinceLastSummary | avg_Data_totalBlobSize showing on Summarize, it should happen after they update their runtime version. See below.

union Office_Fluid_FluidRuntime_*
| where Event_Time > ago(1d)
| where Data_eventName == "fluid:telemetry:Summarizer:Running:Summarize_end"
| where Data_hostScenarioName contains "Whiteboard"
| summarize count() by Data_runtimeVersion

Data_runtimeVersion	count_
0.55.4	81225
0.54.1	1738

Here are the builds where it starts to be populated (0.57.2):
union Office_Fluid_FluidRuntime_*
| where Event_Time > ago(1d)
| where Data_eventName == "fluid:telemetry:Summarizer:Running:Summarize_end"
| where isnotnull(Data_opsSinceLastSummary)
| summarize count() by Data_runtimeVersion

Data_runtimeVersion	count_
0.58.1001	16769
0.58.2002	559
0.58.2001	39
0.57.2	14

#4 -The large numbers comes from the fact we store the time in ms.
So 21678480 is approx = 6 hours (361 mins)

Ex.
union Office_Fluid_FluidRuntime_*
| where Event_Time > ago(7d)
| where Data_eventName == "fluid:telemetry:Summarizer:Running:Summarize_end"
| extend WB = Data_hostScenarioName contains "Whiteboard"
| extend timeSinceLastSummaryHours = iff(isnull(Data_timeSinceLastSummary ), 0.0, Data_timeSinceLastSummary / (1000* 3600))
| summarize count(), round(avg(timeSinceLastSummaryHours)) , round(avg(Data_timeSinceLastAttempt)), round(avg(Data_opsSinceLastSummary)), avg(Data_totalBlobSize), avg(Data_duration)
by Data_eventName, WB

Data_eventName	WB	count_	avg_timeSinceLastSummaryHours	avg_Data_timeSinceLastAttempt	avg_Data_opsSinceLastSummary	avg_Data_totalBlobSize	avg_Data_duration
fluid:telemetry:Summarizer:Running:Summarize_end	0	1609287	4	14801757	28	79123.0775024632	2542.43151532325
fluid:telemetry:Summarizer:Running:Summarize_end	1	885106	3	11025799	NaN	NaN	4380.5799587846

	</BODY></HTML>

[vladsud]

Hey @NicholasCouri , any chance you can provide some insights here on what was wrong with # 4 item from above?
Also can we create some query (filter data) to look at valid data based on earlier available telemetry?
It would be useful to poke at data earlier if there is any way to do so, without waiting for changes to propagate through the system

[NicholasCouri]

@vladsud - Working on it