fankes/beszel
2
0
Fork 0
mirror of https://github.com/fankes/beszel.git synced 2025-10-19 17:59:28 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

enable gpu monitoring by default

Browse Source
This commit is contained in:
Henry Dollman
2024-11-12 18:13:57 -05:00
parent 2e8ac98924
commit b1db450e00
2 changed files with 7 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
beszel/internal/agent/agent.go
View File

@@ -76,12 +76,10 @@ func (a *Agent) Run(pubKey []byte, addr string) {
a.dockerManager = newDockerManager(a) a.dockerManager = newDockerManager(a)
// initialize GPU manager // initialize GPU manager
if os.Getenv("GPU") == "true" { if gm, err := NewGPUManager(); err != nil {
if gm, err := NewGPUManager(); err != nil { slog.Debug("GPU", "err", err)
slog.Warn("GPU", "err", err) } else {
} else { a.gpuManager = gm
a.gpuManager = gm
}
} }
// if debugging, print stats // if debugging, print stats

6
beszel/internal/agent/gpu.go
View File

@@ -57,7 +57,7 @@ func (gm *GPUManager) collectNvidiaStats() error {
} }
// Use a scanner to read each line of output // Use a scanner to read each line of output
scanner := bufio.NewScanner(stdout) scanner := bufio.NewScanner(stdout)
buf := make([]byte, 0, 64*1024) // 64KB buffer buf := make([]byte, 0, 8*1024) // 8KB buffer
scanner.Buffer(buf, bufio.MaxScanTokenSize) scanner.Buffer(buf, bufio.MaxScanTokenSize)
for scanner.Scan() { for scanner.Scan() {
line := scanner.Bytes() line := scanner.Bytes()
@@ -120,7 +120,7 @@ func (gm *GPUManager) startAmdCollector() {
// collectAmdStats runs rocm-smi in a loop and passes the output to parseAmdData // collectAmdStats runs rocm-smi in a loop and passes the output to parseAmdData
func (gm *GPUManager) collectAmdStats() error { func (gm *GPUManager) collectAmdStats() error {
cmd := exec.Command("/bin/sh", "-c", "while true; do rocm-smi --showid --showtemp --showuse --showpower --showproductname --showmeminfo vram --json; sleep 3.7; done") cmd := exec.Command("/bin/sh", "-c", "while true; do rocm-smi --showid --showtemp --showuse --showpower --showproductname --showmeminfo vram --json; sleep 4.3; done")
// Set up a pipe to capture stdout // Set up a pipe to capture stdout
stdout, err := cmd.StdoutPipe() stdout, err := cmd.StdoutPipe()
if err != nil { if err != nil {
@@ -132,7 +132,7 @@ func (gm *GPUManager) collectAmdStats() error {
} }
// Use a scanner to read each line of output // Use a scanner to read each line of output
scanner := bufio.NewScanner(stdout) scanner := bufio.NewScanner(stdout)
buf := make([]byte, 0, 64*1024) // 64KB buffer buf := make([]byte, 0, 8*1024) // 8KB buffer
scanner.Buffer(buf, bufio.MaxScanTokenSize) scanner.Buffer(buf, bufio.MaxScanTokenSize)
for scanner.Scan() { for scanner.Scan() {
var rocmSmiInfo map[string]RocmSmiJson var rocmSmiInfo map[string]RocmSmiJson