Drupal Migration — Performance Monitor

Bottom line — Spectre_5 is healthy overall, with the last 24 hours averaging in the 600–700ms band and apdex mostly 0.75–0.82, aside from a sharp overnight latency spike around 02:00–03:00 CDT. The trailing week held steady at 590–720ms with request volume ranging from ~26K to ~210K/day and error rates under 1% on all but one day. D10 (Drupal 10, live since Jul 4) is tracking almost identically to D9 on latency and apdex, with a slightly lower error rate — no regression from the upgrade so far.

• Overnight latency anomaly: the 02:06 and 03:06 CDT buckets spiked to ~2020–2046ms (vs. a normal 500–700ms), and apdex cratered to 0.59 and 0.53 — the worst apdex readings in the 24h window. Error rate stayed low during this window (0.09–0.14%) and traffic was light (~650–1,100 req), so this reads as a latency-only event (e.g. a cold cache, GC pause, or backend contention), not an error storm.
• Jul 12 saw an elevated daily error rate of 0.82%, roughly 2–3x the surrounding days (0.25–0.42%), worth a quick look at that day's logs even though volume was on the lower side (~64.5K requests).
• D10 early read: 695ms avg response and 0.78 apdex are in line with D9 (699ms / 0.77) and D8 (679ms / 0.79); error rate improved slightly to 0.41% from D9's 0.42%. No signs of a Drupal 10 upgrade regression through this first two-week window.
• Monthly trend: apdex has drifted down gradually from the 0.82–0.83 range in mid-2025 to 0.77–0.79 through 2026 even as monthly request volume grew from ~2.5–2.8M to ~3.3–3.7M — consistent with more load on the same infrastructure rather than a step-change regression.
• Daily volume during the last 24h followed the expected diurnal pattern, peaking at 18.2K req/hr around 14:06 CDT and bottoming out near 654–737 req/hr in the overnight hours, which is exactly when the latency spike above occurred.

Method & caveats

• Cutovers. Drupal 9 / PHP 8 went live May 23, 2026; Drupal 10 Jul 4, 2026 — marked on every chart that spans them.
• Drupal 10 is an early read (~5 days). Its phase card uses the Jul 4–9 window directly.
• Recent spike. Jul 8 shows an isolated response jump (≈2× the prior day), coinciding with elevated reports/progress-API latency that day — an operational blip, not a migration regression.
• Source & retention. New Relic timeslice metrics (HttpDispatcher, Apdex, Errors/all), ~13-month retention — the full arc back to mid-2025 is covered. Percentiles (p95/p99) live in 8-day event data and are reported in the point-in-time reviews, not here.
• Refresh. Regenerated daily by a scheduled routine that re-queries New Relic and redeploys to this same URL.