highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-11-11 00:47:21 +01:00
Code Issues Packages Projects Releases Wiki Activity

test: harden scheduler tests (#12662)

Browse Source 
* test: harden scheduler tests

This removes reschedDelay which was stale code, and adds
a new configurable timeout for the waitForVRAMRecovery so
tests can now set the timeout to be very short to avoid the
scheduler getting stuck and hitting a test timeout.

* test: tune tests for partial loads

Give stress tests more time when the model is split between CPU/GPU
This commit is contained in:
Daniel Hiltgen
2025-10-17 08:56:44 -07:00
committed by GitHub
parent 270679932f
commit 68e04c7ff8
10 changed files with 195 additions and 143 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
integration/utils_test.go
View File

@@ -743,6 +743,13 @@ func skipUnderMinVRAM(t *testing.T, gb uint64) {
// Skip if the target model isn't X% GPU loaded to avoid excessive runtime
func skipIfNotGPULoaded(ctx context.Context, t *testing.T, client *api.Client, model string, minPercent int) {
gpuPercent := getGPUPercent(ctx, t, client, model)
if gpuPercent < minPercent {
t.Skip(fmt.Sprintf("test requires minimum %d%% GPU load, but model %s only has %d%%", minPercent, model, gpuPercent))
}
}
func getGPUPercent(ctx context.Context, t *testing.T, client *api.Client, model string) int {
models, err := client.ListRunning(ctx)
if err != nil {
t.Fatalf("failed to list running models: %s", err)
@@ -772,12 +779,10 @@ func skipIfNotGPULoaded(ctx context.Context, t *testing.T, client *api.Client, m
cpuPercent := math.Round(float64(sizeCPU) / float64(m.Size) * 110)
gpuPercent = int(100 - cpuPercent)
}
if gpuPercent < minPercent {
t.Skip(fmt.Sprintf("test requires minimum %d%% GPU load, but model %s only has %d%%", minPercent, model, gpuPercent))
}
return
return gpuPercent
}
t.Skip(fmt.Sprintf("model %s not loaded - actually loaded: %v", model, loaded))
t.Fatalf("model %s not loaded - actually loaded: %v", model, loaded)
return 0
}
func getTimeouts(t *testing.T) (soft time.Duration, hard time.Duration) {