Resize planner web formation to a single Standard-2X dyno

Offen
  • Stimmt etwas nicht?
#3,006 2 Kommentare 0 Reaktionen 0 zugewiesene Personen Auf GitHub ansehen

Maintainer antworten meist innerhalb von 1 Tag

Dieses Issue hat noch niemand übernommen.

Bewertung

Schwierigkeit
4/5
Geschätzter Aufwand
3-5 Tage
Anfängerfreundlichkeit
18/100
Issue-Typ
Feature
Klarheit
Klar beschrieben
Aktivitätsstatus
Aktiv
Tech-Stack
rails, ruby
Bereich
devops, infrastructure

Rechercherichtung

Review the Heroku formation change and runtime metrics described in the issue, then monitor the single Standard-2X dyno through the two-week evaluation window, including at least two workshop evenings. Compare memory, swap, load, latency, and errors with the stated fail signals and outcome criteria. Done means recording whether the trial passes or fails and whether the Standard-1X trial should be reconsidered.

Vom Indexierungsmodell aus dem Issue-Text verfasst.

Beschreibung

Resize planner web formation to a single Standard-2X dyno

Tracks the web dyno right-sizing experiment: 2× Standard-2X ($100/month) → 1× Standard-2X ($50/month), applied 2026-10-11. The evaluation window is two weeks and must cover at least two workshop evenings. The next step down (a single Standard-1X, $25/month) is on hold until this trial shows the memory headroom in practice.

Context

Heroku runtime metrics from the log drain (7 days, 2026-10-04 → 10-11, recorded while the formation was 2× Standard-2X) showed both web dynos far below the limits of a single Standard-2X. Standard-2X provides 1 GB RAM and a 2× CPU share per dyno.

Preboot is enabled (#2962), so a single dyno gets rolling deploys: the old process keeps serving until the new one finishes booting.

Before (7 days on 2× Standard-2X, per dyno)

Metric Value
Traffic ~11k requests/day; per-minute p99 56
load_avg_1m p99 0.10–0.14, max 0.64
memory_total p50 316–320 MB, p95 372–422 MB, max 403–472 MB
memory_rss p50 285–320 MB, max 351 MB
Swap nonzero in 40% of samples, host-driven; one host moved 138 MB to swap on 10-08 while RSS held 288 MB
R14 / R15 0 occurrences
Latency (p95 of view+db runtime) 50–103 ms daily, 4 errors in ~77k requests

Response-time work this quarter (#2886–#2890 and follow-ups) brought p95 from 788/732 ms to 50–103 ms daily; those changes cut allocations and CPU, not the process memory baseline.

Change

2026-10-11 ~11:11 UTC: heroku ps:scale web=1:standard-2x. Formation is now web=1:Standard-2X; console, rake, release, and worker stay at 0. Delayed jobs are unaffected by the resize: the Heroku Scheduler runs them on one-off dynos (#2942).

Drain evidence at the switch: web.2 received SIGTERM and exited cleanly (status 143) while web.1 kept serving.

After (first ~3 hours — preliminary)

The remaining dyno was fresh (it had cycled that morning), so these numbers are warmup-biased and exclude any evening peak:

Metric Value
memory_total p50 288 MB, max 303 MB
Swap 0% of samples
load_avg_1m p99 0.21, max 0.63

What to watch (fail signals)

  • Any R14 (memory quota exceeded) in web logs.
  • memory_total sustained above 800 MB, or p95 above 700 MB.
  • Swap p95 above 50 MB (swap itself is host-driven; treat trends, not single samples).
  • load_avg_1m p99 above 1.5 during 18:00–22:00 UTC evening peaks.
  • 5xx rate above 10/day or p95 latency above 300 ms.
  • Formation drift: any web dyno other than web.1 appearing in metrics.

Non-memory risks of a single dyno: a crash or hang has no sibling to take traffic until Heroku auto-restart (seconds to minutes); maintenance restarts are a full outage window.

Outcome criteria

Pass: memory settles below ~70% of quota (≈700 MB) with zero R14 across the evaluation window, and latency and error rates stay unchanged from the before window. That pass also yields real single-dyno memory data at the full traffic level.

A steady-state memory_total below ~250 MB would also reopen the case for trying Standard-1X (512 MB quota); below that mark, schedule a guarded one-week trial. Baseline memory is structural (boot, first-touch warmup, slow accumulation), and the caching PRs (#2978, #2964, #2954) show why allocation work did not lower it: request-path allocations are transient, so they cut CPU time and latency, not the process floor.

Fail: any fail signal above. Revert is instant: heroku ps:scale web=2:standard-2x.

Vorherrschende Sprache
Ruby
Sterne
106
Forks
208
Ø Merge
19 Std. 54 Min.
Gemergte PRs (30 T.)
53

Entwicklungsumgebung

Erste Schritte

  1. Lesen Sie das ganze Issue und danach den Beitragsleitfaden des Projekts.
  2. Schreiben Sie ins Issue, dass Sie es übernehmen — das erspart doppelte Arbeit.
  3. Forken Sie das Repository und arbeiten Sie in einem Branch.
  4. Öffnen Sie einen Pull Request, der die Issue-Nummer nennt.

Mehr aus codebar/planner

Alle Issues in codebar/planner

Ähnliche Issues

Weitere Issues zu Ruby

Neue Issues direkt in Ihr Postfach

Eine kurze Übersicht über anfängerfreundliche GitHub-Issues.