TGI queue-size gauge
text-generation-inference · gauge
Fields
id
metric-exposures/w3m-text-generation-inference-tgi-queue-sizeMetric name
tgi_queue_size
Metric type
gauge
Unit
requests
Exposure
prometheus
Endpoint
:9000/metrics
Enabled by default
yes
notes
TGI main, router/src/server.rs plus backends/v3/src/{backend,queue,lib}.rs, read 2026-10-04. On by default: the Prometheus recorder is built unconditionally with PrometheusBuilder and an HTTP listener on prometheus_port, default 9000, so nothing has to be opted into. Label keys are method (prefill|decode) on the batch families, err on tgi_request_failure, and reason on tgi_batch_concat - there are no model or engine labels, so every series is fleet-wide. The waiting-queue depth - TGI's answer to the running/waiting pair. There is no separate running-request gauge; batch shape is visible through tgi_batch_current_size instead.
Notes
TGI main, router/src/server.rs plus backends/v3/src/{backend,queue,lib}.rs, read 2026-10-04. On by default: the Prometheus recorder is built unconditionally with PrometheusBuilder and an HTTP listener on prometheus_port, default 9000, so nothing has to be opted into. Label keys are method (prefill|decode) on the batch families, err on tgi_request_failure, and reason on tgi_batch_concat - there are no model or engine labels, so every series is fleet-wide. The waiting-queue depth - TGI's answer to the running/waiting pair. There is no separate running-request gauge; batch shape is visible through tgi_batch_current_size instead.