Kata Containers tracing status update (2021-05-25)

kata-containers-tracing-status.md

Background

Chelsea and I have been looking at adding end-to-end openTelemetry tracing support to Kata Containers.

Components

With Kata 2.x's simpler architecture, we only need tracing support in:

• hypervisor (ideally)
• runtime
• agent

Note: Not considering hypervisor tracing at the present time.

Status

Runtime

Good progress but still need to a set of consistent and standarised tracing span tags.

Agent

"Beset by problems" (mostly resolved).

---

Agent problems (a selection)

---

• Rebase hell
  • Fast pace of change (good and bad).
  • Testing tracing changes is slow.
  • Lots and lots and lots and lots of rebasing...
  • Rewrites for API changes and async agent changes.

---

• Conversion of VSOCK exporter and trace forwarder to async (done)

---

• Agent shutdown (done)
  • This is only required for tracing.
  • Required a lot of refactoring and we hit a number of bugs on the way.
  • However, shutdown PR landed a code is now much cleaner.
  • ... and the agent can be shut down!

---

• Agent shutdown needs a test (99% done)
  • Debugging tracing problems is very hard.
  • Need to know the agent is ending gracefully.
  • Has to be fairly elaborate.
  • Took a long time to write and test.
  • PR raised.
  • Still not passing in CI env ;(

---

• issues with rust tracing crates (done)
  • A world of pain :-)
  • The rust tracing landscape is still in a state of flux.

---

• Problems with span hierarchy (in progress)
  • Some code cannot be traced without major invasive surgery.
  • Therefore, we register a global tracer.
  • BUT, the means the span hierarchy is difficult to get right.
  • This is not fully resolved yet.

---

Plan

• Raise a foundational agent tracing PR this week
  • Code is not that useful currently
    (due to the span hierarchy issue).
  • However, worth landing this now:
    by landing the basics, we can ratchet up the support iteratively (aka I can avoid wasting lots of time constantly rebasing and re-testing!)
• Send tracing summary to the mailing list.

Request

• "Call to arms"
• Developing and testing tracing is very time consuming.
• Need help from community to speed up progress.
• Volunteer now! ;)
• Ideally, we could use "follow the sun" to speed landing this features.

---

Further details

See the new GitHub project.

---

Questions?

Presented today at the Kata Containers Architecture Committee meeting.

To view as HTML, I ran this:

$ infile="kata-containers-tracing-status.md"
$ outfile="/tmp/tracing.html"
$ pandoc -s --metadata date="25 May 2021" --metadata author="James O. D. Hunt" --metadata title="Kata Containers tracing update" -f markdown -t revealjs -o "$outfile" "$infile" -V revealjs-url=https://unpkg.com/[email protected]/
$ "$BROWSER" "$outfile"

Doc referenced in Kata Mailing List message: http://lists.katacontainers.io/pipermail/kata-dev/2021-May/001934.html.