Skip to main content

v1.0.144

Configuration status tools in MCP​

Agents can now check how Causely is deployed and configured through the Causely MCP server, using two new tools: get_agent_versions and get_notification_configurations. Before, this information was only available on the Agents and Notifications pages in the Causely portal. get_agent_versions returns the mediator, Beyla, and per-host agent versions installed in each cluster, and can list only the clusters that are behind the latest release. get_notification_configurations returns where notifications are sent, which clusters process them, and which filters decide what gets sent. This lets an agent confirm that an update reached every cluster, or explain why a notification was or was not sent.

get_integration_status now also reports, for each Prometheus scraper, how many alert rules are configured and how many alerts are active. This tells apart a cluster whose rules are quiet from a cluster with no rules at all.

Learn more

Zone and region outage diagnosis on GCP​

Causely now diagnoses a cloud zone or region outage once, at the zone or region, instead of once per affected node. Before, Causely had no zone or region entities, so an outage shared by many nodes appeared as a separate diagnosis on each node. Now Causely builds zone and region entities from your nodes and reads the GCP status feed for active compute, Kubernetes, and networking incidents. When the feed reports an incident, a Malfunction diagnosis on the zone or region explains the nodes that are down. This support covers GCP only.

Cluster outages are also recognized earlier. A cluster with two or more nodes down is now diagnosed as one Cluster Malfunction. Before, more than half of the nodes had to be down, and smaller outages produced one diagnosis per node.

Learn more about zones · Learn more about regions

Connection burst detection for non-HTTP services​

Causely now detects reconnect storms on services whose clients connect over raw TCP or WebSocket, such as chat or game services. Before, Client Connection Disruption could only be identified from a burst of HTTP or RPC requests, so a reconnect storm on these services went undiagnosed. Now a Connection Burst Signal fires when a service's rate of new client connections rises above its upper threshold, and Causely traces it to a Client Connection Disruption diagnosis on that service.

You do not need to set the threshold. Causely learns it from the service's own connection rate. To get started, map a connection open rate metric to the service through Prometheus.

Learn more

Minor Improvements​

  • Regex notification filters: Namespace and Cluster notification filters now support REGEX and NOT REGEX, next to IN and NOT IN. Use a pattern such as ^(org|ps)- to match namespaces that are created dynamically.
  • Exclusion filters on missing fields: Notification exclusion filters (not_equals, not_in, not_regex) now pass when an Issue has no value for the field. Before, a namespace exclusion also dropped every node and cluster level Issue. Existing exclusion filters will start delivering Issues that lack the field.
  • SLO status in notifications: The Impacted SLOs section of a notification now shows the correct status for violated and at-risk SLOs. Before, every SLO was reported as Normal.
  • Alertmanager tenant header: Alertmanager notifications can now send an optional X-Scope-OrgID header, for multi-tenant Alertmanager deployments.
  • Faster authentication rejection detection: HTTP 401 Signals on services now activate after 5 minutes instead of 10.
  • Health status in MCP: Signals that are informational, such as Version Changed, no longer mark an entity as degraded in MCP health results or in entity severity.
  • Reliability delta: reliability_delta now reports a warning only when a metric increased, and no longer flags memory that is growing but still below its earlier average as a regression. The output also states how metrics were handled when the version change is older than the stored metric history.
  • Team health in MCP: team_health now matches team names the same way as other tools, including partial matches, and bases its summary on Issues. It returns a clear error when no teams are defined.
  • Log output in MCP: get_logs now truncates long log lines and groups lines by entity, so large log streams fit in an agent's context.
  • Issue lookups in MCP: get_issue_details now returns a specific error when it is called with a diagnosis ID instead of an Issue ID.
  • Alert mapping: Alert names and keywords are now matched regardless of separators, so queue_depth, queue-depth, and queue depth behave the same. Container alerts are mapped to a container from the namespace and pod labels alone.
  • Disk full diagnosis: When a service logs disk full errors and the disk it mounts reports high usage, Causely now produces one Disk Full diagnosis on the service instead of two separate diagnoses.
  • Home: The MCP Inspector is now one of the "Where to go next" cards on Home, and a connected agent is the main action. The welcome dialog and the active Issues section were redesigned.
  • Agent host mounts: The Causely agent now mounts host paths with HostToContainer propagation. This fixes volumes, such as Ceph RBD, staying in use on a node after their pod was rescheduled.