sudo systemctl daemon-reexec
sudo systemctl daemon-reload
sudo systemctl enable bm-agent.service
sudo systemctl start bm-agent.service
sudo systemctl stop bm-agent.service
sudo systemctl status bm-agent.service
sudo systemctl restart bm-agent.service
sudo journalctl -u bm-agent
sudo journalctl -u bm-agent -f
sudo journalctl -u bm-agent -n 300 -f
sudo su - bm-agent
sudo systemctl status bm-scheduler.service
sudo systemctl restart bm-scheduler.service
sudo systemctl stop bm-scheduler.service
sudo journalctl -u bm-scheduler -n 300 -f
sudo su - bm-scheduler
sudo systemctl status bm-monitor.service
sudo systemctl restart bm-monitor.service
sudo systemctl stop bm-monitor.service
sudo journalctl -u bm-monitor -n 300 -f
sudo su - bm-monitor
sudo systemctl status bm-autotuner.service
sudo systemctl restart bm-autotuner.service
sudo systemctl stop bm-autotuner.service
sudo journalctl -u bm-autotuner -n 300 -f
sudo su - bm-autotuner
sudo rm /etc/systemd/system/bm-agent.service
sudo docker exec -it vllm-tpu tail -f /workspace/vllm_log.txt
# create user
sudo useradd -m -s /bin/bash bm-agent
# give docker permission
sudo usermod -aG docker bm-agent
# delete
sudo userdel -r bm-agent
# "login" to this user.
sudo su - bm-agent
buid a local image with local changes
# 1. go to the vllm folder.
# 2. make the changes.
# 3. do git commit.
# run the script scripts/build_and_push_local_image.sh
gcloud secrets create bm-agent-hf-token \
--replication-policy="automatic"
--project=$GCP_PROJECT_ID
echo -n "hf_your_actual_token" | \
gcloud secrets versions add bm-agent-hf-token \
--data-file=- \
--project=$GCP_PROJECT_ID
after resizing the disk
run this to make sure system use it.
sudo resize2fs /dev/nvme0n2
Create instance
gcloud spanner instances create $GCP_INSTANCE_ID \
--description="vllm benchmark run record." \
--nodes=1 \
--project=$GCP_PROJECT_ID \
--config=regional-$REGION
Create database
gcloud spanner databases create $GCP_DATABASE_ID \
--project=$GCP_PROJECT_ID \
--ddl-file=database/vllm_bm.sdl \
--instance=$GCP_INSTANCE_ID
Delete database
gcloud spanner databases delete $GCP_DATABASE_ID \
--instance=$GCP_INSTANCE_ID \
--project=$GCP_PROJECT_ID
Create pubsub
# create topic v6e-1
gcloud pubsub topics create vllm-bm-queue-v6e-1 \
--project="$GCP_PROJECT_ID"
# create agent subscription
gcloud pubsub subscriptions create vllm-bm-queue-v6e-1-agent \
--project="$GCP_PROJECT_ID" \
--topic="vllm-bm-queue-v6e-1" \
--ack-deadline=600
# create topic v6e-4
gcloud pubsub topics create vllm-bm-queue-v6e-4 \
--project="$GCP_PROJECT_ID"
# create agent subscription
gcloud pubsub subscriptions create vllm-bm-queue-v6e-4-agent \
--project="$GCP_PROJECT_ID" \
--topic="vllm-bm-queue-v6e-4" \
--ack-deadline=600
# create topic v6e-8
gcloud pubsub topics create vllm-bm-queue-v6e-8 \
--project="$GCP_PROJECT_ID"
gcloud pubsub subscriptions create vllm-bm-queue-v6e-8-agent \
--project="$GCP_PROJECT_ID" \
--topic="vllm-bm-queue-v6e-8" \
--ack-deadline=600
Delete pubsub
gcloud pubsub topics delete vllm-bm-queue-v6e-1 \
--project=YOUR_PROJECT_ID
gcloud pubsub topics delete vllm-bm-queue-v6e-8 \
--project=YOUR_PROJECT_ID
Give permission to service account.
gcloud projects add-iam-policy-binding $GCP_PROJECT_ID \
--member="<service account>" \
--role="roles/pubsub.publisher"
gcloud projects add-iam-policy-binding $GCP_PROJECT_ID \
--member="<service account>" \
--role="roles/pubsub.subscriber"