Performance

These are measurements from one machine, not limits. We took them on 7 October 2026 with:

• a single-node k3s cluster in colima, with 4 CPUs and 4 GiB of memory, on a 16 GB Apple-silicon MacBook Air;
• KubeGlass built from main that day, on the same laptop;
• Chromium.

The cluster, KubeGlass and the browser shared the laptop, so each number includes the others’ load.

A list of 3,000 objects

3,000 ConfigMaps, shown in one list:

What	Result
Counts	The same as kubectl
First rows, one namespace	324 ms
First rows, all namespaces, nothing cached	3.4 s, with the laptop under heavy load
Scrolling	60 frames a second; 95% of frames took 16.7 ms or less
A keystroke in the filter	32 ms median, 48 ms at most
Sorting	83 to 114 ms
Selecting all 3,001 rows	Under 80 ms
Browser memory (JavaScript heap)	12 to 33 MB, flat over 5 idle minutes

Changes while you watch

20 changes a second to the objects in that list, for 60 seconds:

• No frames were dropped; the longest took 16.8 ms.
• j and k moved the selection in 40 ms (median).
• The 1,200 changes arrived in 546 WebSocket messages.
• The list’s count matched kubectl 8 seconds after the last change.

A busy log

A pod writing 200 lines a second, followed for 3 minutes:

• The view keeps the latest 3,000 lines.
• Browser memory stayed between 12 and 20 MB and didn’t grow.
• The page’s main thread was busy 7 to 10% of the time.

The server

• KubeGlass used 37 to 130 MB of memory across these tests.
• It held one connection to the API server per cluster.

First load

• With KubeGlass on the same laptop, the page showed content in about 0.2 s.
• Over an emulated link with 80 ms of latency and 10 Mbit/s, a spinner showed at 0.1 to 0.25 s and the app at 1.2 to 1.4 s, after 27 to 48 script requests. That was a build of the change, merged on 8 October, that loads the first screen as one script; before it, the page stayed blank until 1.9 to 2.5 s and made 90 to 134.

What we haven’t tested

• Clusters with more than one node.
• Large clusters. Headlamp’s 0.44.0 release notes say it works beyond 30,000 pods; we haven’t run KubeGlass on a cluster anywhere near that size.
• Browsers other than Chromium.

If you run KubeGlass on a bigger cluster, we’d like to hear how it does: open an issue with the cluster’s size and what you saw. For memory in a cluster, see High memory use.