Summary of Diag WG Discussion - 2017/10/05 (Node Interactive)

[mike-kaufman]

Diagnostic WG had a break-out discussion at Node Interactive 2017. This is a summary of the disccusion from Friday afternoon, and is a proposed agenda for next WG meeting.

Please chime in if anything is missing/incorrect. :)

1. Logistics of WG meetings

   • @joshgav was driving these, but he's been on leave.
   • Next Steps:
     • @mike-kaufman & @mrkmarron to sync w/ Josh this week.

2. Async-Context

   • Good discussion on "what is Async Context" .
   • Consensus was we should first close on high level concepts, terminology and use cases
   • some use cases mentioned:
     • long stacks
     • post-mortem analysis - i.e., what was the async path to some point in program execution.
     • understanding details of each async-op on some async-path - i.e., how long spent blocked on IO, how long in event queue, total CPU cycles on path...)
       - others?
     • Next Steps:
       • @mrkmarron to post issue w/ link to paper defining async semantics in javascript
       • @jasnell - do some translation from "academic speak" to "something more accessible"
       • discussion to happen on that issue.

3. Tracing API

   • Node has a tracing trace_events API now (namely Chrome's), but it isn't being fully leveraged by node core
   • Chrome's API may be moving out into separate library, so other libs can take a dep on this
   • Need to understand how tracing trace_events API fits w/ alternative VMs (i.e., node-chakracore)
   • Next Steps:
     • TBD/Discuss at WG meeting

4. Docs/How-Tos

   • missing/non-existent docs now on how to do diagnostic tasks
   • Next Steps:
     • TBD/Discuss at WG meeting

5. Unhandled promise rejections

   • at least 2 PRs open on this
   • not clear if there is "correct" behavior
   • not clear if consensus on what "least sucky" behavior is
   • Next Steps:
     • TBD/Discuss at WG meeting

[AndreasMadsen]

By tracing API do we mean trace_events?

PS: I really dislike the word "tracing" it has too many meanings. Even in our own summaries, I think we need to be more precise.

[jkrems]

@AndreasMadsen Yep, this was about trace_events.

[mrkmarron]

Thanks for the summary notes Mike. I just opened issue #107 for discussion and work around the async-context item.

[AndreasMadsen]

Are we going to have a Hangout meeting about this in the near future? I would like to catch up on the discussions since I couldn't be there.

[mike-kaufman]

Are we going to have a Hangout meeting about this in the near future?

That's the intention. What are the general considerations for scheduling something? (e.g., times that work, how much lead time, other things?)

[mike-kaufman]

FYI - I went ahead and scheduled the next diag WG meeting. See issue here: #108.

[mike-kaufman]

closing this. Minutes are being checked into wg-meetings folder per PR #122