Design: managed metrics collection - #363

Open

dennis-upbound wants to merge 40 commits into

modelplaneai:mainfrom

dennis-upbound:design/metrics

Commits on Sep 3, 2026

Add design for managed metrics collection
dennis-upbound
committed
fca42a4
Nest metrics under observability and address review scope

dennis-upbound
and
claude
committed
c3eebfd
Reframe metrics around always-on collection and central aggregation

dennis-upbound
and
claude
committed
3b19a8f
Scope metrics to per-cluster collection plus a discoverable URL

dennis-upbound
and
claude
committed
3aa8e35
Switch metrics to cluster-wide PodMonitors and state the proposal

dennis-upbound
and
claude
committed
1b32797
Aggregate metrics up to a Modelplane-level view

dennis-upbound
and
claude
committed
5b2fe69
Lock the collector to OpenTelemetry and specify engine capture and normalization

dennis-upbound
and
claude
committed
a1d5e84
Add sequence-length and disaggregation signals, and cluster scheduler metrics

dennis-upbound
and
claude
committed
a9aef40
Stress-test cleanup: managed kube-scheduler, planner framing, error row

dennis-upbound
and
claude
committed
85fd668
Resolve the open questions into decisions

dennis-upbound
and
claude
committed
1e536f1
Scope scaling out and make the mapping registry Crossplane-native

dennis-upbound
and
claude
committed
fe8dc8c
Clarify the mapping registry as a first-class MetricMapping resource

dennis-upbound
and
claude
committed
fec0871
Stress-test cleanup: shared label not registry, throughput row, selector-scoped rename

dennis-upbound
and
claude
committed
ccc2e08
Address the metrics review: connectivity, no mandatory store, EPP agnosticism

dennis-upbound
and
claude
committed
c1a040b
Roll up in the collector, drop the in-memory Prometheus

dennis-upbound
and
claude
committed
35f2002
Cleanup pass: remove stale store references, align roll-up lists

dennis-upbound
and
claude
committed
d4910e7
Record what has landed of the MetricMapping proposal
dennis-upbound
committed
60956da
Record why a MetricMapping selects by label and not by metric name
dennis-upbound
committed
d3d843f
Correct the passthrough claim and the implementation status
dennis-upbound
committed
5798cc6
Reconcile with the Dynamo serving stack
dennis-upbound
committed
bb87ccd
Stamp the engine label instead of asking a user for it
dennis-upbound
committed
8adb769

Commits on Sep 8, 2026

Commit to OpenTelemetry, and say how the series get home
dennis-upbound
committed
7560daa
Cut the counting back out
dennis-upbound
committed
17255e7
Open the engine type, and let the pull fallback exist
dennis-upbound
committed
deb3c9f
Draft the user-facing page, and fix what drafting it exposed
dennis-upbound
committed
1a11a42
Full read-through
dennis-upbound
committed
563766c
Move the docs draft into docs, as a Hugo draft
dennis-upbound
committed
f189afe

Commits on Sep 11, 2026

Support one transport, and name the destination for telemetry

dennis-upbound
and
claude
committed
0c28181

Commits on Sep 14, 2026

Export to the destination directly, and collect from gateway-only clusters

dennis-upbound
and
claude
committed
3384205
Give the right reason for not using the control plane, and follow it

dennis-upbound
and
claude
committed
838e170
Sharpen why the control plane can't host this, and answer the gateway question

dennis-upbound
and
claude
committed
985fdb9
Decide the questions this left open

authored and
Dennis Ramdass
committed
a35053e
Borrow the destination's shape instead of inventing one

authored and
Dennis Ramdass
committed
eb22c8a
Let the OpenTelemetry Operator run the collector

authored and
Dennis Ramdass
committed
de3f520
Read it through after the last three revisions

authored and
Dennis Ramdass
committed
5c8f0a6
Bring the docs draft back in line with the design

authored and
Dennis Ramdass
committed
f545d25
Cut what a reviewer doesn't have to approve

authored and
Dennis Ramdass
committed
d13712d
Say what a hosted control plane is, not whose

authored and
Dennis Ramdass
committed
a22f0a5
Drop the last of the vendor voice

authored and
Dennis Ramdass
committed
2c38024
Answer Christopher's review

authored and
Dennis Ramdass
committed
f2b3a36