From a66dda5ae816a6a8936645fe0520cb4dc6354137 Mon Sep 17 00:00:00 2001
From: Bohumir Zamecnik <bohumir.zamecnik@gmail.com>
Date: Mon, 22 Oct 2018 12:01:13 +0200
Subject: [PATCH] Install as a service via systemd (Docker cannot be used).

---
 DOCKER.md           | 30 +++++++++++++++++++
 Makefile            | 12 ++++----
 README.md           | 71 ++++++++++++++++++++++++++++++++-------------
 nvgpu-agent.service | 15 ++++++++++
 4 files changed, 102 insertions(+), 26 deletions(-)
 create mode 100644 DOCKER.md
 create mode 100644 nvgpu-agent.service

diff --git a/DOCKER.md b/DOCKER.md
new file mode 100644
index 0000000..6dbeca4
--- /dev/null
+++ b/DOCKER.md
@@ -0,0 +1,30 @@
+## Installation and running via Docker
+
+<dev>For easier deployment of the agent apps we can use Docker.</del>
+
+> `psutils` cannot see process details (user, creation time, command) on the host
+OS - by definition os container this is separated. This is a design fault of the
+dockerized solution and I'm not sure if it can work at all.
+See https://github.com/rossumai/nvgpu/issues/2.
+
+It needs [nvidia-docker](https://github.com/NVIDIA/nvidia-docker) installed.
+
+```bash
+# build the image
+docker build -t nvgpu .
+
+# run CLI
+nvidia-docker run --rm nvgpu nvl
+
+# run agent
+nvidia-docker run --rm -p 1080:80 nvgpu
+
+# run the master with agents specified in ~/nvgpu_master.cfg
+nvidia-docker run --rm -p 1080:80 -v $(pwd)/nvgpu_master.cfg:/etc/nvgpu.cfg nvgpu
+
+open http://localhost:1080
+```
+
+You can set the containers for automatic startup with `--restart always` option.
+
+Note: Docker containers have some hash as hostname (it's not the host machine hostname).
diff --git a/Makefile b/Makefile
index a72b08f..560fe14 100644
--- a/Makefile
+++ b/Makefile
@@ -31,9 +31,9 @@ docker_build:
 
 docker_run_nvl:
 	nvidia-docker run --rm nvgpu nvl
-
-docker_run_agent:
-	nvidia-docker run --rm -p 1080:80 nvgpu
-
-docker_run_master:
-	nvidia-docker run --rm -p 1080:80 -v $(pwd)/nvgpu_master.cfg:/etc/nvgpu.cfg nvgpu
+#
+#docker_run_agent:
+#	nvidia-docker run --rm -p 1080:80 nvgpu
+#
+#docker_run_master:
+#	nvidia-docker run --rm -p 1080:80 -v $(pwd)/nvgpu_master.cfg:/etc/nvgpu.cfg nvgpu
diff --git a/README.md b/README.md
index 8025359..3400331 100644
--- a/README.md
+++ b/README.md
@@ -21,15 +21,23 @@ status in a web application.
 
 ## Installing
 
+For a user:
+
+```bash
+pip install -U nvgpu
 ```
-pip install nvgpu
+
+or to the system:
+
+```bash
+sudo -H pip install -U nvgpu
 ```
 
 ## Usage examples
 
 Command-line interface:
 
-```
+```bash
 # grab all available GPUs
 CUDA_VISIBLE_DEVICES=$(nvgpu available)
 
@@ -60,7 +68,7 @@ $ nvl
 
 Python API:
 
-```
+```python
 import nvgpu
 
 nvgpu.available_gpus()
@@ -97,7 +105,7 @@ Agents can also display their status by default.
 
 ### Agent
 
-```
+```bash
 FLASK_APP=nvgpu.webapp flask run --host 0.0.0.0 --port 1080
 ```
 
@@ -118,37 +126,60 @@ AGENTS = [
 ]
 ```
 
-```
+```bash
 NVGPU_CLUSTER_CFG=/path/to/nvgpu_master.cfg FLASK_APP=nvgpu.webapp flask run --host 0.0.0.0 --port 1080
 ```
 
 Open the master in the web browser: http://node01:1080.
 
-## Installation and running via Docker
+## Installing as a service
 
-For easier deployment of the agent apps we can use Docker.
-
-It needs [nvidia-docker](https://github.com/NVIDIA/nvidia-docker) installed.
+On Ubuntu with `systemd` we can install the agents/master as as service to be
+ran automatically on system start.
 
 ```bash
-# build the image
-docker build -t nvgpu .
+# create an unprivileged system user
+sudo useradd -r nvgpu
+```
 
-# run CLI
-nvidia-docker run --rm nvgpu nvl
+Copy [nvgpu-agent.service](nvgpu-agent.service) to:
 
-# run agent
-nvidia-docker run --rm -p 1080:80 nvgpu
+```bash
+sudo vi /etc/systemd/system/nvgpu-agent.service
+```
 
-# run the master with agents specified in ~/nvgpu_master.cfg
-nvidia-docker run --rm -p 1080:80 -v $(pwd)/nvgpu_master.cfg:/etc/nvgpu.cfg nvgpu
+Set agents to the configuration file for the master:
 
-open http://localhost:1080
+```bash
+sudo vi /etc/nvgpu.conf
 ```
 
-You can set the containers for automatic startup with `--restart always` option.
+```python
+AGENTS = [
+         # direct access without using HTTP
+         'self',
+         'http://node01:1080',
+         'http://node02:1080',
+         'http://node03:1080',
+         'http://node04:1080',
+]
+```
 
-Note: Docker containers have some hash as hostname (it's not the host machine hostname).
+Set up and start the service:
+
+```bash
+# enable for automatic startup at boot
+sudo systemctl enable nvgpu-agent.service
+# start
+sudo systemctl start nvgpu-agent.service 
+# check the status
+sudo systemctl status nvgpu-agent.service
+```
+
+```bash
+# check the service
+open http://localhost:1080
+```
 
 ## Author
 
diff --git a/nvgpu-agent.service b/nvgpu-agent.service
new file mode 100644
index 0000000..76609ae
--- /dev/null
+++ b/nvgpu-agent.service
@@ -0,0 +1,15 @@
+[Unit]
+Description=NVGPU agent
+
+[Service]
+User=nvgpu
+Group=nvgpu
+Environment=NVGPU_CLUSTER_CFG=/etc/nvgpu.conf
+Environment=FLASK_APP=nvgpu.webapp
+Environment=FLASK_HOST=0.0.0.0
+Environment=FLASK_PORT=1080
+ExecStart=/bin/bash -c "/usr/local/bin/flask run --host $FLASK_HOST --port $FLASK_PORT"
+Restart=always
+
+[Install]
+WantedBy=multi-user.target