Page MenuHomePhabricator

Benchmark Shellbox
Closed, ResolvedPublic

Event Timeline

The lilypond file I picked/generated is pretty simple:

\header {
	tagline = ##f
\version "2.18.2"
\language "nederlands"
\score {

\relative c' { f d f a }
\layout { }
	\midi {
		\context { \Score tempoWholesPerMinute = #(ly:make-moment 100 4) }

\paper {
	indent = 0\mm

I started using ab to fire off requests with increasing concurrency, at c = 25 is when pyball depooled once othe k8s nodes so I paused it as I tweak.

Screenshot 2021-07-09 at 16-13-19 Envoy Telemetry (k8s) - Grafana.png (1×3 px, 387 KB)

lilypond_lilypond_c20.png (600×800 px, 7 KB)

lilypond_lilypond_c25.png (600×800 px, 7 KB)

Shocking, more replicas increases concurrency.

I need to set up a dashboard so I can monitor pod memory/CPU usage while benchmarking to see how beneficial it will be to increase that.

thanks! Forked to

Note that deployment="$namespace" didn't work for me, I had to switch it to kubernetes_namespace="$namespace"

Legoktm claimed this task.

I think we're mostly set with the current configuration. Benchmarking and real usage shows that for the most part Shellbox is sitting idle, but the current capacity is adequate to handle spikes when multiple pages with scores are edited/re-parsed in quick succession.