fix(edxapp): source Granian concurrency per install from stack config

[blarghmatey]

What are the relevant tickets?

tk-stage-3-adopt-new-granian-defaults-for-mitxonlin-16de40, project wp-granian-configuration-overhaul-expose-blocking-t-3debc2. Follows #5607.

Description (What does this do?)

k8s_resources.py is shared by four Open edX installs (mitx, mitx-staging, mitxonline, xpro), so a concurrency value written there is written for all four at once. That is how the 2026-08-26 CMS regression reached mitxonline Studio. Per #5607's post-mortem, the rollout cut per-pod backpressure 8x (2 workers x 64 to 1 x 16), APISIX p95/p99 reached the 60s histogram ceiling, and Tempo placed 30-59s ahead of a Django span that itself took 0.2-1.1s. granian_connections_active for mitxonline CMS pinned at 15-16 through that window and has been back at 50 (p99, 3d) since the rollback.

#5607 mitigated it by restoring the holding pins in shared code. That is correct for mitxonline and collateral damage for the other two: mitx and mitx-staging CMS have run 1 worker x 8 threads since 2026-08-24T17:18Z and would be dragged back to 2 workers on their next deploy.

Both webapps now read edxapp:k8s_granian.{lms,cms} from stack config, alongside the existing k8s_replicas and k8s_resources. The key is required rather than defaulted, so an install's concurrency is readable in its own stack file and a new install has to state a value rather than inherit whatever was last written into shared code.

install	LMS	CMS
mitx, mitx-staging	1 x 8, backpressure 16	1 x 8, backpressure 16
mitxonline, xpro	2 x 32, backpressure 64	2 x 32, backpressure 64

Every value matches what is live as of 2026-08-28, read from the container args on applications-production and residential-production.

How can this be tested?

pulumi preview against each Production stack:

• mitxonline.Production: 238 unchanged, zero changes. The holding pins are preserved exactly.
• mitx.Production: 3 Deployment updates (--disable-prefetch, from fix(edxapp): add --disable-prefetch to all celery workers to fix KEDA scaling #5598) plus 2 RDS alarm deletions. No Granian arg diff. Previewing the same stack against origin/main instead shows a 4th Deployment update, on CMS: [7]: "1" => "2" (workers), --runtime-mode mt reinserted, blocking-threads 8 => 32, backpressure 16 => 64, workers-max-rss 1843 => 921. That is the collateral revert this PR prevents.
• mitx-staging.Production: a Vault mount option update, 3 --disable-prefetch, 2 RDS alarm deletions. No Granian arg diff.
• xpro.Production: previews cleanly and renders --blocking-threads 32 --backpressure 64. Its 127 changes are dominated by the pre-existing ~6-month drift on that stack (it still runs the hand-rolled xpro-production-edxapp-{cms,lms}-webapp Deployments), unrelated to this PR.

pre-commit run over the changed files passes, mypy included.

Additional notes

mitx CMS does touch its backpressure ceiling of 16, for 52 minutes out of 3 days (granian_connections_active >= 16, 1m resolution), with a peak granian_blocking_queue of 5 and zero container restarts. mitx-staging CMS: 33 minutes, peak queue 0. Healthy, but the ceiling is the signal to watch if authoring traffic on either grows.

Raising mitxonline off the holding pins still needs concurrency sizing that accounts for burst shape, plus a scaling or alerting signal that sees connection saturation -- CPU and request rate did not.

🤖 Generated with Claude Code

https://claude.ai/code/session_01HgKWpUsasvv1pLGBDhyBo5

[Copilot]

Pull request overview

Moves Open edX LMS and CMS Granian concurrency from shared code into per-stack configuration, preserving install-specific production capacity.

Changes:

• Requires separate LMS and CMS Granian settings.
• Configures all 12 edxapp stacks with appropriate concurrency.
• Documents the rollout and operational rationale.

Reviewed changes

Copilot reviewed 14 out of 14 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
Pulumi.xpro.QA.yaml	Adds xpro QA concurrency pins.
Pulumi.xpro.Production.yaml	Adds xpro production concurrency pins.
Pulumi.xpro.CI.yaml	Adds xpro CI concurrency pins.
Pulumi.mitxonline.QA.yaml	Adds mitxonline QA concurrency pins.
Pulumi.mitxonline.Production.yaml	Preserves production holding pins.
Pulumi.mitxonline.CI.yaml	Adds mitxonline CI concurrency pins.
Pulumi.mitx.QA.yaml	Adds CMS settings alongside LMS settings.
Pulumi.mitx.Production.yaml	Preserves tuned mitx concurrency.
Pulumi.mitx.CI.yaml	Adds tuned CMS concurrency.
Pulumi.mitx-staging.QA.yaml	Adds tuned CMS concurrency.
Pulumi.mitx-staging.Production.yaml	Preserves tuned staging concurrency.
Pulumi.mitx-staging.CI.yaml	Adds tuned CMS concurrency.
k8s_resources.py	Sources both webapps’ Granian configuration per stack.
granian-configuration-overhaul.md	Records the per-install design and rollout status.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.