Alerting on Temporal SDK Worker metrics

View Markdown

Temporal SDKs emit metrics from your Worker processes that expose failure modes the Temporal Service cannot see on your behalf: Workflow code that fails on replay, Workers that stop polling, Task slots that never free up, and Local Activities that outlive the Workflow Task heartbeat window.

This page recommends a set of alerts covering those failure modes. It applies to Workers connected to Temporal Cloud and to a self-hosted Temporal Service.

For metric definitions and tag sets, see the Temporal SDK metrics reference. For Worker sizing and tuning, see Worker performance and Worker deployment and performance.

Start with these five

If you are adding Worker alerts for the first time, start here. These five catch the failure modes that stop Workflow Executions outright, and they produce the fewest false positives.

1. All pollers disconnected — Workers have stopped polling.
2. Non-determinism error — Workflow code no longer matches recorded history.
3. gRPC message too large — Executions are being terminated with work lost.
4. Workflow Task schedule-to-start latency elevated — Tasks are backing up.
5. RESOURCE_EXHAUSTED on user-facing operations — starts, Signals, and Updates are being throttled.

Add the rest once these are tuned and quiet.

Choose your thresholds

Every threshold on this page is a starting point, not a service level objective.

A high-throughput Task Queue, a latency-sensitive Namespace, and a bursty batch workload each want different values. Two things to tune per alert:

• The threshold determines what counts as unhealthy. Set it against observed p99 during a known-good period, not against the value listed here.
• The for duration determines how long the condition must hold before the alert fires. Shorter durations detect problems faster and fire more often on transient spikes. Longer durations suppress noise and delay detection.

For alerts on binary conditions — any occurrence of a gRPC status code, a gauge reaching zero — the for duration is doing all the work. Those conditions occur briefly during normal operation, so a for duration that is too short will page you on routine events.

Alerts that fire during normal operation

Several of these conditions are expected in a healthy deployment. Set the for duration long enough to ride them out, and confirm your value against your own deploy cadence:

Event	Alerts it can trigger
Worker deploy or rolling restart	NOT_FOUND on respond operations, all pollers disconnected, Task completions dropped to zero, non-determinism error
Temporal Service upgrade	INTERNAL from the Temporal Service
Normal Workflow completion or termination while a Task is in flight	NOT_FOUND on respond operations
Idle or low-volume Task Queue	Task completions dropped to zero, sticky cache holding zero entries
Scale-down or Worker autoscaling	All pollers disconnected, Task slots exhausted

Recommended alert set

The condition column gives the tag filters that distinguish each alert. Group each alert by namespace and by the tags in its condition, so a firing alert tells you which Namespace, operation, or Task Queue is affected.

Request failures

These fire on gRPC responses from the Temporal Service to your Worker or Client. Triage guidance: SDK request failures.

Failure mode	Metric	Condition	Threshold	for	Default severity
NOT_FOUND on respond operations	request_failure	status_code=NOT_FOUND, operation in RespondWorkflowTaskCompleted, RespondWorkflowTaskFailed, RespondActivityTaskCompleted, RespondActivityTaskFailed	Any occurrence	5m	Critical
NOT_FOUND on Activity heartbeat	request_failure	status_code=NOT_FOUND, operation=RecordActivityTaskHeartbeat	Any occurrence	5m	Warning
RESOURCE_EXHAUSTED on user-facing operations	request_failure	status_code=RESOURCE_EXHAUSTED, operation in StartWorkflowExecution, SignalWithStartWorkflowExecution, SignalWorkflowExecution, UpdateWorkflowExecution, ExecuteMultiOperation	Any occurrence	1m	Critical
RESOURCE_EXHAUSTED on respond operations	request_failure	status_code=RESOURCE_EXHAUSTED, operation in the four respond operations above	Any occurrence	5m	Critical
RESOURCE_EXHAUSTED on poll operations	long_request_failure	status_code=RESOURCE_EXHAUSTED, operation in PollWorkflowTaskQueue, PollActivityTaskQueue	Any occurrence	5m	Warning
UNIMPLEMENTED from the Temporal Service	request_failure	status_code=UNIMPLEMENTED, any operation	Any occurrence	2m	Critical
INTERNAL from the Temporal Service	request_failure	status_code=INTERNAL, any operation	Any occurrence	2m	Critical
Request latency high on user-facing operations	request_latency	operation in the five user-facing operations above	p99 above 2s	5m	Critical

Worker capacity

These fire when Workers stop keeping up with the Task Queue. Triage guidance: SDK Worker capacity.

Failure mode	Metric	Condition	Threshold	for	Default severity
Worker Task slots exhausted	worker_task_slots_available	worker_type in WorkflowWorker, ActivityWorker, LocalActivityWorker	Reaches 0	2m	Critical
All pollers disconnected	num_pollers	poller_type in workflow_task, workflow_sticky_task, activity_task	Reaches 0	5m	Critical
Task completions dropped to zero	request	operation in RespondWorkflowTaskCompleted, RespondActivityTaskCompleted	Rate reaches 0 while the Task Queue has demand	5m	Critical
Workflow Task schedule-to-start latency elevated	workflow_task_schedule_to_start_latency	task_queue	p99 above 5s	5m	Warning
Workflow Task schedule-to-start latency severe	workflow_task_schedule_to_start_latency	task_queue	p99 above 30m	5m	Critical
Activity schedule-to-start latency severe	activity_schedule_to_start_latency	task_queue	p99 above 30m	5m	Critical
Sticky cache holding zero entries under load	sticky_cache_size	Paired with a non-zero Workflow Task rate on the same Worker	Reaches 0	15m	Warning

info

worker_task_slots_available reports meaningful values only with fixed-size slot suppliers. It cannot be used with resource-based slot suppliers — see Slot availability metrics. If your Workers use resource-based tuning, alert on schedule-to-start latency instead and treat it as the signal that Worker capacity is short.

Execution failures

These fire on Workflow and Activity code failing on the Worker. Triage guidance: SDK execution failures.

Failure mode	Metric	Condition	Threshold	for	Default severity
Non-determinism error	workflow_task_execution_failed	failure_reason=NonDeterminismError	Any occurrence	5m	Critical
gRPC message too large	workflow_task_execution_failed	failure_reason=GrpcMessageTooLarge	Any occurrence	1m	Critical
Workflow Task execution failures elevated	workflow_task_execution_failed	failure_reason=WorkflowError	Rate above 10/s	2m	Warning
Workflow Task execution latency high	workflow_task_execution_latency	task_queue, workflow_type	p99 above 10s	5m	Critical
Activity execution failures elevated	activity_execution_failed	activity_type	Rate above 10/s	2m	Warning
Local Activity latency exceeds the heartbeat timeout	local_activity_execution_latency	activity_type	p99 above 30m	5m	Critical

Set up the alerts

Before you can alert on these metrics, your Workers must be emitting them and your monitoring system must be scraping them:

• Temporal Cloud — Set up SDK metrics, then configure Prometheus and Grafana.
• Self-hosted — Monitor Temporal Platform metrics.

Resolve the metric names for your setup

The metric names in the tables above are the base names from the SDK metrics reference. The names you query depend on your SDK and metrics reporter:

• All metrics carry a temporal_ prefix.
• Counters take a _total suffix when scraped through Prometheus: temporal_request_failure_total.
• Histograms take a _seconds_bucket suffix on the underlying bucket series: temporal_workflow_task_schedule_to_start_latency_seconds_bucket.
• Gauges take no suffix. temporal_num_pollers, temporal_worker_task_slots_available, and temporal_sticky_cache_size are gauges. Querying them with _total returns no data.

Tag values also vary. The status_code tag is PascalCase in the Go SDK (NotFound) and UPPER_SNAKE_CASE in the Java and Core SDKs (NOT_FOUND). Some tags are not emitted by every SDK — sticky_cache_size, for example, carries namespace only in the TypeScript and Java SDKs and task_queue only in TypeScript.

Confirm the exact names and tags in your own metrics endpoint before writing queries.

Route alerts by severity

The severity in each table is a default routing suggestion, not a property of the metric. Tune it to your workload.

As a starting point, page on alerts whose condition means Workflow Executions have stopped progressing or are progressing with data loss or duplicate side effects. Send the rest to a review channel.

Two reasons the defaults will not fit every deployment:

• Some Critical rows are latency thresholds or leading indicators rather than confirmed stoppage. Elevated Workflow Task execution latency matters a great deal on a latency-sensitive Namespace and very little on a batch workload where Tasks routinely run long.
• Some Warning rows can still cause duplicate side effects. A NOT_FOUND on Activity heartbeat means the Activity attempt was already timed out and will re-execute from scratch, which is a real problem if that Activity is not idempotent.

Several of these alerts have causal relationships. Task slots exhausted drives pollers to zero, which drives schedule-to-start latency up, which drives Task completions to zero. When several fire together, the triage pages identify which is the root cause and which are symptoms.