Before you connect
This release supports approved endpoints reachable from a self-hosted OpenCluster deployment, including self-hosted Loki and Grafana Cloud Loki. A deployment administrator must approve each destination and its network access. Prefer HTTPS and a read-only token. TLS certificates are verified, redirects are refused, and private endpoints must be directly reachable; Loki requests do not use a Relay. Admins and Editors can revise Integration configuration. A configuration-only edit retains the existing token, and subsequent verification or queries use it with the revised URL, authentication and tenant settings. Trust configuration administrators accordingly, approve changes before applying them, and verify again after an edit. OpenCluster does not enforce a destination allowlist. A multi-tenant OpenCluster Cloud offering needs a separate egress and destination policy before accepting arbitrary customer URLs.Create the Integration
As an Admin, create a Loki Integration with these configuration values:
For Grafana Cloud, use its Loki URL and user, select
basic, and supply a read-only access
policy token. For a self-hosted gateway, select the authentication mechanism it documents.
No arbitrary headers or additional authentication mechanisms are supported.
The authenticated product API accepts a Bearer configuration such as:
POST /api/v1/integrations with the authentication described in the
API overview. Include token when required;
it is never returned in configuration reads. To replace it, use
PATCH /api/v1/integrations/{integration} with the complete non-secret configuration
and the new token. Omitting token during an edit preserves the existing credential.
Creation verifies actual read access with a one-minute log query, {job=~".+"}, limited
to one entry. A valid empty response also proves access; it does not prove that a job
label or matching logs exist. A readiness check alone is insufficient. Successful
verification makes the log Tool eligible through the normal access checks.
Investigate service logs
Ask: “What errors did checkout log during the latency spike? Compare them with its metrics and the deployment.” Use a known service label and a focused expression:loki_query_logs; label discovery is not exposed. Supply known selectors from your
instrumentation or available context rather than assuming all deployments use app.
The result includes the actual expression, applied window, direction and limit,
timestamped excerpts with stream labels, and an
empty, data or incomplete outcome.
Timestamps retain nanosecond precision. Entries are ordered across streams before the
local limit is applied. URLs, credentials and tenant IDs are selected from Integration
configuration, never Tool arguments.
Limits and interpretation
- The query covers at most the latest 24 hours of the requested Investigation interval. A narrowed window is marked incomplete. Empty and disjoint intervals are refused.
- Reaching the requested entry limit is conservatively marked incomplete, even when the source may have exactly that many entries. Longer lines are excerpted to 2000 characters and marked incomplete. Query a narrower interval or selector when more detail is needed.
- HTTP calls have a 10-second timeout and respect Investigation cancellation. Responses are limited to 2 MiB after decompression; normalized results must fit 256 KiB. Oversized results fail explicitly rather than sending unlimited logs to the model.
- Only the read-only
GET /loki/api/v1/query_rangeendpoint is used. Metric results, malformed streams and invalid or out-of-window entry timestamps fail explicitly. Structured metadata and query statistics are not included in excerpts. - Empty results do not prove health, absence of an error, or causation. Log retention, ingestion delays, permissions and query selection affect which entries are available.
- Arbitrary LogQL is evaluated by Loki. Time bounds are not a complete historical-access or query-cost policy; apply retention, tenant isolation and query limits at the source. Select log fields carefully because upstream logs can themselves contain sensitive data.