Skip to content

[serve.llm][dashboard] Configure aggregation interval #56591

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 4 commits into from
Sep 19, 2025
+69 −29
Merged

[serve.llm][dashboard] Configure aggregation interval #56591

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
046b403
Configure aggregation interval
eicherseiji Sep 16, 2025
3b886aa
Merge branch 'master' into llm-dash-configurable-interval
eicherseiji Sep 16, 2025
3c2e611
Fix units
eicherseiji Sep 16, 2025
ea9209f
Merge branch 'master' into llm-dash-configurable-interval
eicherseiji Sep 17, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
58 changes: 29 additions & 29 deletions python/ray/dashboard/modules/metrics/dashboards/serve_llm_dashboard_panels.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -17,11 +17,11 @@
unit="tokens/s",
targets=[
Target(
expr='sum by (model_name, WorkerId) (rate(ray_vllm:request_prompt_tokens_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by (model_name, WorkerId) (rate(ray_vllm:request_prompt_tokens_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="Prompt Tokens/Sec - {{model_name}} - {{WorkerId}}",
),
Target(
expr='sum by (model_name, WorkerId) (rate(ray_vllm:generation_tokens_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by (model_name, WorkerId) (rate(ray_vllm:generation_tokens_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="Generation Tokens/Sec - {{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -33,27 +33,27 @@
Panel(
id=2,
title="vLLM: Time Per Output Token Latency",
description="Time per output token latency in milliseconds.",
unit="ms",
description="Time per output token latency.",
unit="s",
targets=[
Target(
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P99 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P95 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P90 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P50 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='(sum by(model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='(sum by(model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:time_per_output_token_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="Mean - {{model_name}} - {{WorkerId}}",
),
],
Expand Down Expand Up @@ -85,27 +85,27 @@
Panel(
id=5,
title="vLLM: Time To First Token Latency",
description="P50, P90, P95, and P99 TTFT latency in milliseconds.",
unit="ms",
description="P50, P90, P95, and P99 TTFT latency.",
unit="s",
targets=[
Target(
expr='(sum by(model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='(sum by(model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="Average - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P50 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P90 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P95 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId)(rate(ray_vllm:time_to_first_token_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P99 - {{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -121,23 +121,23 @@
unit="s",
targets=[
Target(
expr='sum by(model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))\n/\nsum by(model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_count{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="Average - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.5, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P50 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.9, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P90 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.95, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P95 - {{model_name}} - {{WorkerId}}",
),
Target(
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])))',
expr='histogram_quantile(0.99, sum by(le, model_name, WorkerId) (rate(ray_vllm:e2e_request_latency_seconds_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])))',
legend="P99 - {{model_name}} - {{WorkerId}}",
),
],
Expand Down Expand Up @@ -177,7 +177,7 @@
unit="Requests",
targets=[
Target(
expr='sum by(le, model_name, WorkerId) (increase(ray_vllm:request_prompt_tokens_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(le, model_name, WorkerId) (increase(ray_vllm:request_prompt_tokens_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="{{le}}",
template=TargetTemplate.HEATMAP,
),
Expand All @@ -195,7 +195,7 @@
unit="Requests",
targets=[
Target(
expr='sum by(le, model_name, WorkerId) (increase(ray_vllm:request_generation_tokens_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(le, model_name, WorkerId) (increase(ray_vllm:request_generation_tokens_bucket{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="{{le}}",
template=TargetTemplate.HEATMAP,
),
Expand All @@ -213,7 +213,7 @@
unit="Requests",
targets=[
Target(
expr='sum by(finished_reason, model_name, WorkerId) (increase(ray_vllm:request_success_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(finished_reason, model_name, WorkerId) (increase(ray_vllm:request_success_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="{{finished_reason}} - {{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -229,7 +229,7 @@
unit="s",
targets=[
Target(
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_queue_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_queue_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="{{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -245,11 +245,11 @@
unit="s",
targets=[
Target(
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_decode_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_decode_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="Decode - {{model_name}} - {{WorkerId}}",
),
Target(
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_prefill_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_prefill_time_seconds_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="Prefill - {{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -265,7 +265,7 @@
unit="none",
targets=[
Target(
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_max_num_generation_tokens_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]))',
expr='sum by(model_name, WorkerId) (rate(ray_vllm:request_max_num_generation_tokens_sum{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]))',
legend="{{model_name}} - {{WorkerId}}",
),
],
Expand All @@ -281,7 +281,7 @@
unit="percentunit",
targets=[
Target(
expr='increase(ray_vllm:gpu_prefix_cache_hits_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s]) / increase(ray_vllm:gpu_prefix_cache_queries_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[30s])',
expr='increase(ray_vllm:gpu_prefix_cache_hits_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval]) / increase(ray_vllm:gpu_prefix_cache_queries_total{{model_name=~"$vllm_model_name", WorkerId=~"$workerid", {global_filters}}}[$interval])',
legend="GPU: {{model_name}} - {{WorkerId}}",
),
],
Expand Down
40 changes: 40 additions & 0 deletions python/ray/dashboard/modules/metrics/dashboards/serve_llm_grafana_dashboard_base.json
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -89,6 +89,46 @@
"$__all"
]
}
},
{
"name": "interval",
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

will users actually know what this means and why they should customize it?

Alternatively, you can use the grafana built in variables $__rate_interval or $__interval which is set by grafana based on what timerange they select.

I think in the grafana UI, you can explicitly override the intervals as well and it will apply it to all the promql if you use the $__interval variable.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Discussion offline: tried $__interval but no results are shown. $__interval may resolve to interval too small to capture two data points.

Advantage to configurable interval: lower interval can be helpful if developing on a head node where metrics are scraped more frequently. And longer can be helpful if you want to easily calculate e.g. the average over an entire benchmark

"label": "Interval",
"type": "custom",
"hide": 0,
"includeAll": false,
"multi": false,
"options": [
{
"selected": true,
"text": "30s",
"value": "30s"
},
{
"selected": false,
"text": "1m",
"value": "1m"
},
{
"selected": false,
"text": "5m",
"value": "5m"
},
{
"selected": false,
"text": "10m",
"value": "10m"
},
{
"selected": false,
"text": "15m",
"value": "15m"
}
],
"current": {
"selected": true,
"text": "5m",
"value": "5m"
}
}
]
},
Expand Down