1073458 - Kibana query to report the number of calls per hour daily

Status: RESOLVED WONTFIX

(Cloud Services :: Operations: Metrics/Monitoring, task)

Description

[Romain Testard [:RT]]

As a product manager I want to know daily the number of successful calls which occurred per 60 minute slot so that I know the service peak times.

Comment 1

[Alexis Metaireau (:alexis)]

We don't have the success information in the server right now. In order to have that, the clients will need to send the server a notification once the call is finished. We currently don't have any websocket connection at this point in time (it's already finished).

We could have a way to let the server know, but that will mean the server will know more information about the calls going on. The other way to do that is to have the client directly report to a metrics server.

I'm ccing kparlante here since she's the one handling metrics.

Comment 2

[Alexis Metaireau (:alexis)]

Do we want something like this? https://github.com/mozilla-services/loop-server/graphs/punch-card

Comment 3

[Romain Testard [:RT]]

The only issue I see with this is that it does not tell me what the real numbers are.
More like https://github.com/mozilla-services/loop-server/graphs/commit-activity (top part with orange rectangles). This will allow seeing real numbers of calls per 60 minute window

Comment 4

[Katie Parlante]

Assuming we use the same definition of "success" as the other call related metrics (really successful call setup), we should be able to use websocket.summary. Again, not blocked on getting the machinery going, just want to make sure the websocket.summary data is right.

Thanks for the examples, perhaps we can get metrics team feedback on visualization.

Comment 5

[Katie Parlante]

fyi, here's a histogram of calls/hour for the last month: https://kibana.shared.us-west-2.prod.mozaws.net/#dashboard/temp/rRykaGfmQa2fjVgwDxfBDw

Comment 6

[Romain Testard [:RT]]

I like that!
NI Adam in case he has further thoughts as he is interested in that graph too.

Comment 7

[Adam Roach [:abr]]

This looks very nice.

I don't think I'd said much about the calls-per-hour graph. What I did comment on earlier -- and we'd need data from TokBox to implement this -- is a call duration distribution graph. Something roughly like this:

  100000 |
         |               #####
C  10000 |               #####
o        |        #####  #####
u   1000 |        #####  #####
n        |        #####  #####  #####
t    100 | #####  #####  #####  #####
         | #####  #####  #####  #####  #####
      10 | #####  #####  #####  #####  #####
  -------+------------------------------------
         |  < 1    1-5    5-10   10-60  >60
         |     Length of call (in minutes)

Comment 8

[Romain Testard [:RT]]

I updated the US to include room data.

FYI 
* call duration spread bug: 1073451
* room duration spread bug: 1106056

Comment 9

[Wesley Dawson [:whd]]

Reopen if necessary.