fix(status): bound the stacked-area forward-fill to each series' cadence

[dawsontoth]

Stacked on #1587 — review that one first. Base auto-retargets to stage when it merges.

#1587 bounded last-observation-carry-forward for cross-node gauge sums in the pipeline. StackedAreaChart has a second, independent forward-fill at the render layer, and it was unbounded: it merges every series onto the union of x positions and carries each series' last-seen value across the positions it did not report, with no staleness limit.

Why #1587's bound can't reach this

In Stack by: Node, TrafficByTypeRenderer remaps the spec's dimension to 'node', so each series holds exactly one node and that node is present in every bucket the series has. Since pipeline carry-forward never invents bucket times, it is a structural no-op in that mode — nothing but this merge bridges those bands.

That is both why node mode never showed the #1576 dives, and why an idle node's band was carried to the right-hand edge of the window. Core emits mqtt-connections only while the count is nonzero (if (numberOfConnections > 0) in server/mqtt.ts), so "no row" is exactly what an idle node looks like — the same overstatement #1587 fixed, one layer up.

What changed

The fill stays; it is load-bearing. period: 0 gauge rows snap onto the spec's 60 s fallback lattice, which a 90 s emission cadence beats against, so per-node lattice coverage is ragged and a strict merge draws a shredded stack. What changes is that a band now expires:

STALENESS_INTERVALS × max(median reporting gap, lattice step) + one lattice step

That trailing lattice step is slack the pipeline does not need. The pipeline estimates cadence from raw emission instants; a primitive only ever sees times already snapped onto the lattice, and snapping compresses alternate gaps — a 90 s cadence on a 60 s lattice reports at 0/120/180/300…, so its gaps alternate 120/60 and their median reads 60 or 90 depending on how many samples the window caught, while the widest gap is 120. Without the slack a series can expire on the very beat pattern the snap induced (I hit a zero-margin case before adding it). One step is the most the snap can stretch a gap by, which keeps this horizon at least as generous as the pipeline's.

Cadence is derived from the series' own point spacing rather than threaded through SeriesData: derived metrics (mqtt-traffic-*) and hand-built SeriesData reach this primitive too, so a threaded field would be absent often enough that this derivation would remain the live path anyway.

Past the horizon a band fills 0, not null

• It mirrors the pipeline — dropping an expired node from a crossNode: 'sum' is contributing 0 to that sum.
• It is what a stacked chart means: bands add, so 0 is the neutral element, while a null punches a hole through the stack and drops every band above it.
• A cluster-wide gap still reads as a gap. X positions come only from series that actually reported, so when everything stops the rows stop rather than filling with zeros.

Regression check across every stacked-area panel

Panel	Aggregator	Effect
connections	max / sum	the motivating fix — gauge
database-size	last / sum	same gauge shape
bytes-sent / bytes-received	sum / sum	improves
mqtt-traffic-sent / -received	sum / sum	improves

The additive counters improve for the same reason #1587 excluded them from pipeline carry-forward: a missing bucket for a rate series means no traffic, so expiring to 0 is the honest value where carrying the last throughput forward invented traffic indefinitely.

Observed values are never rewritten — including a genuine drop to 0 — and nothing is back-filled before a series' first point. The ceiling series stays unbounded deliberately: it is a reference line drawn over the stack, not a band in it, with no zero to expire to (a ceiling at 0 reads as a limit of nothing).

csvExport is untouched. It exports raw points with empty cells for gaps and never shared this fill — the export is the data, not the drawing.

Testing

Extracted the merge to mergeStackedRows so it is unit-testable at all: every assertion about rendered geometry in the existing stacked-area suite is it.skip, because neither jsdom nor happy-dom lays Recharts out. Also memoized it in the component, which re-renders on every hover tick.

16 cases in __tests__/primitives/stacked-row-staleness.test.ts, shaped after cross-node-gauge-gaps.test.ts and driven through the real connections / bytes-sent specs and the real pipeline: the staggered-cadence shape actually producing ragged coverage, the pipeline no-op in node mode, horizon expiry, still-bridged-inside-the-horizon, no flicker back, the exact horizon boundary on hand-checkable numbers, observed-value passthrough, no back-fill, y: null handling, no invented rows, and the ceiling exception.

Verified the coverage is not vacuous by reverting the horizon to Infinity — 4 cases fail, then restored.

Gate: 265 files / 1898 tests pass, tsc -b, oxlint, and dprint all clean.

Not browser-verified. Reproducing the artifact needs a genuinely idle node in a live cluster, which can't be summoned on demand, and port 5173 was held by another checkout. Verification is mechanism-level, through the real specs and pipeline.

Refs #1576

🤖 Generated with Claude Code

[gemini-code-assist]

Code Review

This pull request extracts and refactors the merging and forward-filling logic of StackedAreaChart into a separate utility module mergeStackedRows.ts. It introduces a bounded carry-forward mechanism based on a calculated staleness horizon to prevent idle nodes from contributing to the stacked total indefinitely (fixing issue #1576). Comprehensive regression tests are also added in stacked-row-staleness.test.ts. There are no review comments, so I have no additional feedback to provide.

[cb1kenobi]

Reviewed a092573 — no issues found. This PR looks good, nice job!

—
Generated by Barber AI