From: Peter Amstutz <peter.amstutz@curii.com>
Date: Mon, 10 Jan 2022 20:39:26 +0000 (-0500)
Subject: 12630: Tweak nvidia warnings
X-Git-Tag: 2.4.0~115^2~1
X-Git-Url: https://git.arvados.org/arvados.git/commitdiff_plain/5e06ca0b451f36be33396f8e83bdaa4f9d6f74bb

12630: Tweak nvidia warnings

Arvados-DCO-1.1-Signed-off-by: Peter Amstutz <peter.amstutz@curii.com>
---

diff --git a/lib/crunchrun/cuda.go b/lib/crunchrun/cuda.go
index 8282359eab..6efc689bad 100644
--- a/lib/crunchrun/cuda.go
+++ b/lib/crunchrun/cuda.go
@@ -5,15 +5,13 @@
 package crunchrun
 
 import (
-	"fmt"
-	"io"
 	"os/exec"
 )
 
 // nvidiaModprobe makes sure all the nvidia kernel modules and devices
 // are set up.  If we don't have all the modules/devices set up we get
 // "CUDA_ERROR_UNKNOWN".
-func nvidiaModprobe(writer io.Writer) {
+func nvidiaModprobe(writer *ThrottledLogger) {
 	// The underlying problem is that when normally running
 	// directly on the host, the CUDA SDK will automatically
 	// detect and set up the devices on demand.  However, when
@@ -44,7 +42,7 @@ func nvidiaModprobe(writer io.Writer) {
 	nvidiaSmi.Stderr = writer
 	err := nvidiaSmi.Run()
 	if err != nil {
-		writer.Write([]byte(fmt.Sprintf("nvidia-smi error: %v\n", err)))
+		writer.Printf("Warning - nvidia-smi: %v", err)
 	}
 
 	// Load the kernel modules & devices associated with
@@ -65,7 +63,7 @@ func nvidiaModprobe(writer io.Writer) {
 		nvmodprobe.Stderr = writer
 		err = nvmodprobe.Run()
 		if err != nil {
-			writer.Write([]byte(fmt.Sprintf("nvidia-modprobe error: %v\n", err)))
+			writer.Printf("Warning: nvidia-modprobe: %v", err)
 		}
 	}
 }