Found running it through the cluster: - Strata's server starts its engine in a process group of its own, so the agent saw a GPU replica on the CPU and its 22 GB as somebody else's memory. A process the shard started, at any remove, is now the shard's. - An adopted shard's kept compute was sent at once and never again, so a wrong one read again on adoption never reached the control plane: what is sent is sent again when it changes, and a GPU shard kept as the CPU is read again on adoption. - /metrics/prometheus is llama.cpp's names without the KV cache; vLLM's, with it, are /metrics?format=prometheus. The router now sees running 1, waiting 2 under load. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
39 lines
1.2 KiB
Go
39 lines
1.2 KiB
Go
package scrape
|
|
|
|
import (
|
|
"strings"
|
|
"testing"
|
|
)
|
|
|
|
// strataSample is Strata 0.2.7's /metrics?format=prometheus, cut down, as
|
|
// cuda01 served it: vLLM's names beside its own.
|
|
const strataSample = `# TYPE vllm:num_requests_running gauge
|
|
vllm:num_requests_running{model_name="qwen3.8-flash-next"} 1
|
|
# TYPE vllm:num_requests_waiting gauge
|
|
vllm:num_requests_waiting{model_name="qwen3.8-flash-next"} 3
|
|
# TYPE vllm:kv_cache_usage_perc gauge
|
|
vllm:kv_cache_usage_perc{model_name="qwen3.8-flash-next"} 0.25
|
|
# TYPE vllm:prompt_tokens_total counter
|
|
vllm:prompt_tokens_total{model_name="qwen3.8-flash-next"} 4826
|
|
# TYPE vllm:generation_tokens_total counter
|
|
vllm:generation_tokens_total{model_name="qwen3.8-flash-next"} 1979
|
|
# TYPE strata:live_tok_s gauge
|
|
strata:live_tok_s 0
|
|
`
|
|
|
|
func TestStrataIsReadInVLLMsNamesFromItsPrometheusFormat(t *testing.T) {
|
|
m := MappingFor("strata")
|
|
// Plain /metrics is JSON, and /metrics/prometheus has llama.cpp's
|
|
// names without the KV cache: read there, Strata was unmeasured.
|
|
if m.Path != "/metrics?format=prometheus" {
|
|
t.Errorf("read at %q", m.Path)
|
|
}
|
|
s, err := Parse(strings.NewReader(strataSample), m)
|
|
if err != nil {
|
|
t.Fatal(err)
|
|
}
|
|
if s.Running != 1 || s.Waiting != 3 || s.KVCacheUsage != 0.25 {
|
|
t.Errorf("stats %+v", s)
|
|
}
|
|
}
|