perf(state): lean entity model + kwok scale lab - #2
Merged
Conversation
Local high-scale harness (scripts/scale/) using kwok to simulate large clusters with near-zero compute: fake nodes report huge capacity and fake pods reach Running instantly. - up/down: create/delete the kwok cluster (docker runtime) - load: 20 nodes, 20 namespaces, 10,000 pods (+ rs/svc/cm/secrets), tunable via KRUST_* env vars (total = NAMESPACES * DEPLOYS_PER_NS * REPLICAS) - contexts: generate a 20-context kubeconfig (distinct default namespaces) to exercise multi-context / warm-context paths against one apiserver - churn: rolling deployment restarts at a configurable rate - gitignore the generated kubeconfig (embeds kwok cluster certs) Used to capture the stable-release performance baselines. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
Stop retaining the full object JSON for every cached entity. At 10k pods in one context the store held ~482 MB (~48 KB/pod) because ResourceEntity.raw kept the entire serialized object. - Replace ResourceEntity.raw with a compact Extracted payload (node name, container names, ownerReferences, pod resource req/lim, node capacity), computed once at ingest in cluster/extract.rs and dropped. - Fetch the full object on demand for detail/describe/decode/edit via the new ResourceProvider::get_object, cached single-slot (detail_cache). - Pulse aggregate and log fan-in owner mapping now read extracted fields instead of walking raw JSON for every pod. Also adds a hidden headless benchmark mode (--bench, src/ui/app/bench.rs) that boots the real data plane against a cluster and times the project/render/pulse hot paths + RSS, plus the stable-release ROADMAP and the 1.1 design note. Measured on the kwok 10k-pod lab (1 context, release build): RSS 482 MB -> 85.5 MB (5.6x smaller) pulse aggregate 7.62 ms -> 0.36 ms (21x faster) projection/render unchanged (Phase 2.1); initial sync unchanged — per-event to_value remains, addressed in Phase 1.2. Validation: fmt, check, test (71 passed / 3 ignored), release build all green. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
ErfanY
added a commit
that referenced
this pull request
Jun 25, 2026
In the logs pane, `P` (shift+p) toggles between the current and the previous (terminated) container instance's logs — `kubectl logs -p`, for debugging crashlooping containers. Closes gap #2 from the k9s parity audit (4.7). - One-shot fetch: previous logs don't follow and don't auto-reconnect (the container is gone), so the reconnect loop is skipped while previous is active. - Refetches on toggle (clears the buffer); a fresh log view always starts on the current instance. - Status line shows `instance:current|previous`. The PodLogRequest.previous plumbing already existed; this wires the UI to it. Test: toggle flips the flag, resets the buffer, and the title reflects the instance. Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
First slice of the stable-release refactor (readonly-first, 20+ clusters / 10k pods/cluster). Measurement-driven: a kwok scale lab + headless bench, then the biggest memory win.
What's here
Phase 0 — scale lab & measurement
scripts/scale/(kwok): 20 nodes, 20 namespaces, 10,000 pods, 20-context kubeconfig, churn generator. Tunable viaKRUST_*env.--benchmode (src/ui/app/bench.rs) boots the real data plane and times project/render/pulse hot paths + RSS.Phase 1.1 — lean entity model
ResourceEntity.raw); replace with a compactExtractedpayload computed once at ingest (cluster/extract.rs).ResourceProvider::get_object, single-slotdetail_cache.Measured (kwok 10k pods, 1 context, release build)
to_value(Phase 1.2)Validation
fmt · check · test (71 passed / 3 ignored) · release build — all green.
Notes
docs/design/phase-1.1-lean-entity-model.md; tracker:ROADMAP.md.maintriggersrelease.yml, so merge when ready to cut a version.🤖 Generated with Claude Code