Supported Metrics

Warning

Note that CTA Metrics is in its early stages and considered experimental. Published metrics and their attributes are likely to change.

Metric and attribute names follow OpenTelemetry semantic conventions, with CTA-specific prefixes for internal domains (e.g., cta.taped, cta.scheduler).

Info

For how to configure CTA to publish metrics, see Enabling Metrics.

Metrics

Metric Name	Type	Unit	Description	Attributes
db.client.connection.count	UpDownCounter	1	The number of connections that are currently in a state described by the state attribute.	db.namespace db.system.name state
db.client.operation.duration	Histogram	ms	Duration of database client operations.	db.namespace db.system.name
cta.frontend.request.duration	Histogram	ms	Duration the frontend takes to process a request.	event.name
cta.scheduler.operation.duration	Histogram	ms	Duration of a CTA scheduling operation.	cta.scheduler.operation.name
cta.objectstore.lock.acquire.duration	Histogram	ms	Duration taken to acquire an objectstore lock.	lock.type
cta.taped.mount.duration	Histogram	ms	Duration to mount a tape.	cta.transfer.direction
cta.taped.transfer.count	Counter	1	Number of files transferred to/from tape.	cta.transfer.direction
cta.taped.transfer.io	Counter	by	Bytes transferred to/from tape.	cta.transfer.direction

Resource Attributes

Attribute Name	Description
service.namespace	Logical namespace of the service emitting the metric. Equivalent to the instance name in CTA.
service.name	Name of the service emitting the metric (e.g. cta.taped, cta.frontend).
service.version	Version of the service emitting the metric.
service.instance.id	Unique identifier for the specific service instance. Useful when multiple replicas run under the same namespace.
process.title	Title of the process within the service. For cta.taped, this means per-drive.
host.name	Host on which the service is running.
cta.scheduler.namespace	Logical name of the scheduler backend in use (e.g. disk, tape).

Metric Attributes

Attribute Name	Description
db.namespace	Database namespace (schema or logical grouping).
db.system.name	Name of the database system (e.g., postgresql, oracle).
cta.scheduler.operation.name	Name of the CTA scheduling operation (e.g. enqueueArchive, cancelRepack).
cta.frontend.requester.name	Name of the frontend event requester (e.g. user, subsystem, or service calling the API).
cta.transfer.direction	Direction of the transfer (archive or retrieve).
lock.type	Type of lock being acquired in the object store or internal resource (e.g., read, write).
cta.taped.thread_pool.name	Name of the thread pool handling taped operations.
event.name	Name of the event being tracked (e.g., frontend or scheduler event).
error.type	Classification of an error that occurred (e.g., network, timeout, permission_denied).
state	Operational or lifecycle state represented by the metric (e.g., active, queued, failed).
le	Histogram bucket upper bound (“less than or equal” duration in ms).

---

Note on Resource vs Metric Attributes

Resource attributes describe the entity that produced the telemetry (service/process/host). They are not automatically attached to every metric by all backends.

• In Prometheus, resource attributes are exposed as a separate time series named target_info (and related info series). They are not labels on each metric by default.
• If you need resource attributes as labels on every series, either:
• enable resource -> metric label conversion in your OpenTelemetry Collector pipeline, or
• join metrics with target_info in PromQL (e.g., on(...) group_left(...)) at query time.
• Prometheus label keys are sanitized to be valid identifiers (e.g., service.instance.id -> service_instance_id).