Run metric requests in batches of eight workers, isolate endpoint failures,
and preserve CPU and memory history when Sentinel lacks a summary endpoint.
Repair Sentinel tokens before creating workers and bound SSH fetch timeouts.
Reauthorize team servers on refresh and hydration, clear stale metrics after
transfers or deletions, and resolve container metadata for team ID zero.
Use only root filesystem samples when mount fields are present, while
supporting legacy disk data. Add regression coverage for concurrent workers,
access changes, endpoint failures, and disk mapping.
Address review findings on the fleet metrics page.
Data correctness:
- bucket raw Sentinel samples into epoch-aligned windows (5m/30m/2h per range)
and average them, replacing per-server LTTB. Every server now reports the
same bucket timestamps, so the fleet sums add up point by point instead of
zigzagging between servers (and RX+TX really combine)
- keep only the root mount in disk history, matching HasMetrics::getDiskMetrics
- leave sparkline gaps as null instead of 0, so absent samples read as gaps
Robustness:
- normalize containerMetric in mount()/updatedContainerMetric(), so
/metrics?cmetric=foo no longer throws an undefined array key
- include the server uuid in the container row wire:key; system containers
share one Sentinel id across servers
- round the range start to the minute so the 60s history cache can hit
- cap the Sentinel curl with --max-time 10, so one hung request cannot stall
the serial per-server loop
UI:
- pass the spark time axis on first paint, so tooltips no longer show 1970
- show the active series name (RX/TX) in the server network chart tooltip
Refine the /metrics fleet page and align the per-server metrics page with it.
Fleet overview:
- add live sparklines to the CPU/Memory/Disk/Network KPI tiles (shared axis,
capped to ~40 points per sparkline)
- fill the Servers online + Containers tiles (fleet-health bar, hottest
container) so all six tiles read as one system
- scope-aware labels: drop the "Fleet" prefix and helper copy when a single
server is selected
Trend charts:
- hero layout: CPU spans full width, the other four form a 2x2 grid
- apply the same hero layout + skeleton to the per-server page for consistency
Hottest containers table:
- add a rank column and a per-row magnitude bar scaled to the busiest container
- fix a leaked Blade @break directive that rendered literally and forced
horizontal scroll on the Network tab
- keep the value column a fixed width so bars align across rows
Live refresh:
- add analytics-style loading feedback to the range tabs and server filter
- stable wire:key on table rows + width transitions on bars so data glides
on live refresh instead of snapping
Generalize the shared _sparkline partial (optional color/label/format) so it
serves non-traffic metrics without changing analytics behaviour.
Aditya Tripathi
Andras BacsaiandClaude Opus 5.5