Development edition · 16 November 2026. Commands using repository paths, Make targets or sample projects require access to the Quinzee source checkout. No public installer is available. Examples use disposable local environments.
12. Observe the platform and the application separately
Quinzee's platform diagnostics describe the emulator, workers and helpers. Application Insights and Log Analytics store application telemetry through Azure-compatible APIs. An optional external Grafana/OTLP stack observes the platform. These are related but separate pipelines: enabling one does not silently configure the others.
Step 1: check basic health
GET /health is public liveness. GET /ready says Identity and Resource Manager
are loaded; it does not certify that every optional module or external database
is available. In QHQ → Monitoring → Observability → Health, inspect workers,
module state, engine status and heartbeat freshness. An old runtime status can
remain visible after its helper stops; use the heartbeat as well as the last
reconciliation result.
An engine heartbeat older than ten seconds is unavailable. Request handler metrics exclude socket transfer and executor wait, while transport metrics span the parsed HTTP exchange. Queue depth samples cover at most 32 queues with a bounded sampling budget and ten-second cache; a truncated sample is not a complete count of pending work. Use service inventory for exact operational investigation.
Step 2: correlate one request
Open Requests, reproduce one action, and filter by its trace ID. A trace ID links request/operation records; it is not an authentication token. Logs omit headers, queries, bodies and private credential material. Application log messages and custom properties remain application-controlled data, so producers must avoid secrets and personal data themselves. Redaction of known names is not a general sensitive-data detector.
Administrator APIs include /quinzee/v1/observability and
/quinzee/v1/observability/metrics. The latter returns Prometheus text. Both
require an administrator bearer token; do not publish them anonymously to make
scraping convenient.
Step 3: provision and query application monitoring
Compile examples/book-iac/monitoring.bicep into
.local/admin-training/monitoring.json with python3 scripts/deps/bicep.py build
and --outfile, then submit it through QHQ Deployments in your training group.
The default parameters create a Log Analytics workspace and a workspace-backed
Application Insights component. The component needs the workspace capability, so unload the component
module before the workspace module. Configure the application's official exporter
with its returned connection string and supported token credential when local
authentication is disabled. Use 100% sampling for this subset.
Open Workspace logs, enter the customer/workspace ID and a query:
AppTraces | order by TimeGenerated desc | take 20
Use AppRequests, AppDependencies, AppTraces, AppExceptions, AppEvents or AppMetrics. Supported stages are simple where comparisons, project, one-column order by, take/limit, count and summarize count(). Unsupported operators return errors; there is no full KQL engine behind the textbox.
| Resource / API | Configurable values and bounds |
|---|---|
| Workspace | Name/location/tags, PerGB2018 SKU, retentionInDays 30–730 |
| Component | Name/location/tags, web/other application type, immutable workspace binding, DisableLocalAuth |
| Ingestion | Uncompressed JSON object/array/NDJSON, supported envelope types, 1 MiB/256 envelopes, 8 KiB rows |
| Stored data | At most 64 workspaces/components; latest 4,096 rows/workspace in addition to time retention |
| Query | Query and optional timespan; 4,096 characters, 16 stages, at most 1,000 result rows; QHQ displays up to 100 |
| ARM metrics | Supported request/dependency count/failed/duration; Count/Total/Average/Minimum/Maximum; PT1M/PT5M/PT1H/FULL; at most ten names, 30 days and 1,440 buckets |
The 4,096-row cap can evict data much sooner than the configured days. Do not promise a month of audit history based only on retentionInDays. Live metrics, profiling, alerts, diagnostic settings, gzip/sample ingestion and arbitrary cross-workspace queries are unsupported.
Step 4: enable external platform telemetry
The simplest complete local stack is make application-observability-up.
It prepares a separate application environment, certificates, private telemetry
credentials and the Grafana/Collector/Prometheus/Loki/Tempo services. Record the
URLs and paths it reports. For a separately managed cloud, use the normal TLS
arguments plus --otlp-endpoint https://<collector-origin> and
--otlp-ca <collector-ca-file>. The engine helper has matching
QUINZEE_OTLP_ENDPOINT/QUINZEE_OTLP_CA environment values.
The stack Compose settings include QUINZEE_UID/GID, QUINZEE_OBSERVABILITY_DIR
(private persisted data and TLS/auth files), QUINZEE_GRAFANA_PORT (default 13001)
and QUINZEE_OTLP_PORT (14318). Its configuration files are mounted from
infrastructure/observability: collector.yaml routes telemetry, prometheus.yaml
sets collection, loki.yaml and tempo.yaml control stores, grafana.ini and the
provisioning directory set dashboard/datasource behavior. Review those explicit
files when operating the optional third-party stack; they are not quinzee.cloud
JSON members. Keep the healthcheck and authentication/trust mounts intact.
Export is bounded and best effort. Collector failure increments export/drop counters instead of blocking application requests. Check those counters before assuming that an empty dashboard proves no requests occurred. Stop/backup this stack separately from the cloud's own monitoring resource files.