Add GCP wait-start-ssh helper
Browse files- scripts/gcp_wait_start_ssh.sh +105 -0
scripts/gcp_wait_start_ssh.sh
ADDED
|
@@ -0,0 +1,105 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
#!/usr/bin/env bash
|
| 2 |
+
# meshai-a100-trainer: start + SSH hazir olana kadar bekle / tekrar dene.
|
| 3 |
+
# Cloud Shell veya gcloud kurulu herhangi bir makinede:
|
| 4 |
+
# bash scripts/gcp_wait_start_ssh.sh
|
| 5 |
+
# bash scripts/gcp_wait_start_ssh.sh --no-ssh # sadece ac, SSH yapma
|
| 6 |
+
# bash scripts/gcp_wait_start_ssh.sh --cmd 'cd ~/MeshAI_Cloud && ...'
|
| 7 |
+
|
| 8 |
+
set -euo pipefail
|
| 9 |
+
|
| 10 |
+
INSTANCE="${INSTANCE:-meshai-a100-trainer}"
|
| 11 |
+
ZONE="${ZONE:-europe-west4-a}"
|
| 12 |
+
MAX_ATTEMPTS="${MAX_ATTEMPTS:-60}"
|
| 13 |
+
SLEEP_SEC="${SLEEP_SEC:-20}"
|
| 14 |
+
DO_SSH=1
|
| 15 |
+
REMOTE_CMD=""
|
| 16 |
+
|
| 17 |
+
while [[ $# -gt 0 ]]; do
|
| 18 |
+
case "$1" in
|
| 19 |
+
--no-ssh) DO_SSH=0; shift ;;
|
| 20 |
+
--cmd) REMOTE_CMD="$2"; shift 2 ;;
|
| 21 |
+
--zone) ZONE="$2"; shift 2 ;;
|
| 22 |
+
--instance) INSTANCE="$2"; shift 2 ;;
|
| 23 |
+
*) echo "Bilinmeyen arg: $1"; exit 1 ;;
|
| 24 |
+
esac
|
| 25 |
+
done
|
| 26 |
+
|
| 27 |
+
log() { echo "[$(date '+%H:%M:%S')] $*"; }
|
| 28 |
+
|
| 29 |
+
need_gcloud() {
|
| 30 |
+
if ! command -v gcloud >/dev/null 2>&1; then
|
| 31 |
+
echo "HATA: gcloud yok. Bu scripti Cloud Shell'de calistir."
|
| 32 |
+
exit 1
|
| 33 |
+
fi
|
| 34 |
+
}
|
| 35 |
+
|
| 36 |
+
instance_status() {
|
| 37 |
+
gcloud compute instances describe "$INSTANCE" --zone="$ZONE" --format='get(status)' 2>/dev/null || echo "UNKNOWN"
|
| 38 |
+
}
|
| 39 |
+
|
| 40 |
+
ensure_started() {
|
| 41 |
+
local st
|
| 42 |
+
st="$(instance_status)"
|
| 43 |
+
log "Durum: $st"
|
| 44 |
+
if [[ "$st" == "RUNNING" ]]; then
|
| 45 |
+
return 0
|
| 46 |
+
fi
|
| 47 |
+
if [[ "$st" == "STAGING" || "$st" == "PROVISIONING" || "$st" == "STOPPING" ]]; then
|
| 48 |
+
log "Gecis durumu ($st) — bekleniyor..."
|
| 49 |
+
return 1
|
| 50 |
+
fi
|
| 51 |
+
# TERMINATED / STOPPED / SUSPENDED / UNKNOWN
|
| 52 |
+
log "Start deneniyor: $INSTANCE ($ZONE)"
|
| 53 |
+
if gcloud compute instances start "$INSTANCE" --zone="$ZONE"; then
|
| 54 |
+
log "Start komutu OK"
|
| 55 |
+
return 0
|
| 56 |
+
fi
|
| 57 |
+
log "Start basarisiz — tekrar denenecek"
|
| 58 |
+
return 1
|
| 59 |
+
}
|
| 60 |
+
|
| 61 |
+
ssh_ready() {
|
| 62 |
+
# Exit 0 if SSH accepts a trivial command
|
| 63 |
+
gcloud compute ssh "$INSTANCE" --zone="$ZONE" --command="echo SSH_OK" --quiet 2>/dev/null | grep -q SSH_OK
|
| 64 |
+
}
|
| 65 |
+
|
| 66 |
+
need_gcloud
|
| 67 |
+
log "Hedef: $INSTANCE / $ZONE (max $MAX_ATTEMPTS x ${SLEEP_SEC}s)"
|
| 68 |
+
|
| 69 |
+
attempt=0
|
| 70 |
+
while (( attempt < MAX_ATTEMPTS )); do
|
| 71 |
+
attempt=$((attempt + 1))
|
| 72 |
+
log "Deneme $attempt/$MAX_ATTEMPTS"
|
| 73 |
+
|
| 74 |
+
ensure_started || true
|
| 75 |
+
sleep 5
|
| 76 |
+
|
| 77 |
+
st="$(instance_status)"
|
| 78 |
+
if [[ "$st" != "RUNNING" ]]; then
|
| 79 |
+
log "Henuz RUNNING degil ($st) — ${SLEEP_SEC}s"
|
| 80 |
+
sleep "$SLEEP_SEC"
|
| 81 |
+
continue
|
| 82 |
+
fi
|
| 83 |
+
|
| 84 |
+
log "RUNNING — SSH kontrol..."
|
| 85 |
+
if ssh_ready; then
|
| 86 |
+
log "SSH hazir."
|
| 87 |
+
if [[ "$DO_SSH" -eq 0 ]]; then
|
| 88 |
+
log " --no-ssh: cikis (VM acik)"
|
| 89 |
+
exit 0
|
| 90 |
+
fi
|
| 91 |
+
if [[ -n "$REMOTE_CMD" ]]; then
|
| 92 |
+
log "Uzak komut calistiriliyor..."
|
| 93 |
+
gcloud compute ssh "$INSTANCE" --zone="$ZONE" --command="$REMOTE_CMD"
|
| 94 |
+
exit $?
|
| 95 |
+
fi
|
| 96 |
+
log "Interaktif SSH aciliyor..."
|
| 97 |
+
exec gcloud compute ssh "$INSTANCE" --zone="$ZONE"
|
| 98 |
+
fi
|
| 99 |
+
|
| 100 |
+
log "SSH henuz yok — ${SLEEP_SEC}s sonra tekrar"
|
| 101 |
+
sleep "$SLEEP_SEC"
|
| 102 |
+
done
|
| 103 |
+
|
| 104 |
+
log "HATA: $MAX_ATTEMPTS denemede SSH hazir olmadi."
|
| 105 |
+
exit 1
|