metrics: fix application_instances_per_channel disagreeing with UI instance counts

[riyacore404]

application_instances_per_channel metric undercounts/miscounts instances

The nebraska_application_instances_per_channel gauge had three bugs relative
to the rest of the codebase:

1. Dead instances counted forever — the query never filtered on
   last_check_for_updates, so instances that stopped checking in stayed in
   the metric indefinitely instead of aging out like every other
   activity-based query (GetApp, GetApps, GetInstances, etc).
2. Instances with no channel silently dropped — the query used comma
   joins across groups/channel, so any instance whose group had no
   channel assigned (groups.channel_id can be NULL) disappeared from the
   metric entirely rather than being counted.
3. amd64/arm64 channels merged into one series — the query grouped only
   by channel name, so two channels sharing a name across architectures
   (e.g. "stable" for amd64 and arm64) were folded into a single row.

This fixes all three: adds the same freshness filter used elsewhere, switches
to LEFT JOIN for groups/channel so channel-less instances land in an
explicit "no channel" bucket instead of vanishing, and adds arch to the
grouping so per-architecture counts are reported separately.

Fixes #1562

How to use

No config or API changes — the fix is internal to how
GetAppInstancesPerChannelMetrics computes the application_instances_per_channel
Prometheus gauge. Reviewers can validate by scraping /metrics on a
Nebraska instance with some inactive instances and/or a channel-less group
and confirming the counts now match what's shown in the instances UI.

Note: this PR adds a new arch label to the application_instances_per_channel
gauge. This is a breaking change for any existing dashboards/alerts that group
by application/version/channel alone — they'll now see one series per
(channel, arch) pair instead of one per channel. Flagging this explicitly in
case maintainers would rather split the label change into a separate PR from
the activity-window/no-channel fixes.

Testing done

Added/updated unit tests in backend/pkg/api/metrics_test.go covering all
three bugs, run against sample_data.sql via the local Postgres test service:

## Test Results

```bash
cd backend
docker compose -f ./docker-compose.test.yaml up -d postgres
NEBRASKA_DB_URL="postgres://postgres:nebraska@127.0.0.1:8001/nebraska_tests?sslmode=disable&connect_timeout=10"
go test ./pkg/api/... -run 'TestGetAppInstancesPerChannelMetrics' -v
docker compose -f ./docker-compose.test.yaml down

Output

=== RUN TestGetAppInstancesPerChannelMetrics
--- PASS: TestGetAppInstancesPerChannelMetrics (0.06s)
=== RUN TestGetAppInstancesPerChannelMetricsExcludesStaleInstances
--- PASS: TestGetAppInstancesPerChannelMetricsExcludesStaleInstances (0.07s)
=== RUN TestGetAppInstancesPerChannelMetricsIncludesGroupsWithoutChannel
--- PASS: TestGetAppInstancesPerChannelMetricsIncludesGroupsWithoutChannel (0.07s)
=== RUN TestGetAppInstancesPerChannelMetricsSeparatesArch
--- PASS: TestGetAppInstancesPerChannelMetricsSeparatesArch (0.07s)
PASS
ok github.com/flatcar/nebraska/backend/pkg/api 0.342s

All targeted tests pass successfully.

Note: I wasn't able to run the full code-checks (golangci-lint) target locally due to environment constraints. Happy to address any lint feedback from CI.

- [ ] Changelog entries added in the respective `changelog/` directory (user-facing change, bug fix, security fix, update)
- [ ] Inspected CI output for image differences: `/boot` and `/usr` size, packages, list files for any missing binaries, kernel modules, config files, kernel modules, etc.

<!-- For coreos-overlay ebuild modifications that include a CROS_WORKON_COMMIT bump, did you bump too the ebuild revision ? -->

[Copilot]

Pull request overview

Fixes the nebraska_application_instances_per_channel Prometheus gauge to align with UI instance counts by making the underlying query consistent with the rest of the “active instance” logic and by preventing instances from being dropped/merged incorrectly.

Changes:

• Add an activity-window filter (last_check_for_updates within validityInterval) to stop counting stale instances.
• Switch to LEFT JOIN + COALESCE so instances in groups without a channel are still counted (in an explicit “no channel” bucket).
• Add arch to the grouping and metric labels to avoid merging same-named channels across architectures, and update unit tests accordingly.

Reviewed changes

Copilot reviewed 4 out of 4 changed files in this pull request and generated 1 comment.

File	Description
backend/pkg/metrics/metrics.go	Adds arch label handling when exporting the gauge.
backend/pkg/api/types/metrics.go	Extends the metric row type with an Arch field (with -1 sentinel for no-channel).
backend/pkg/api/metrics_test.go	Adds/updates tests to cover stale-instance filtering, channel-less groups, and arch separation.
backend/pkg/api/internal/dbreads/metrics.go	Rewrites the SQL query to apply the freshness filter, use LEFT JOIN, and group by arch.

---

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