Add idle culler

docs/index.rst

@@ -123,6 +123,7 @@ Topic guides provide in-depth explanations of specific topics.
    topic/tljh-config
    topic/authenticator-configuration
    topic/escape-hatch
+   topic/idle-culler
 
 
 Troubleshooting

docs/topic/idle-culler.rst

@@ -0,0 +1,86 @@
+.. _topic/idle-culler:
+
+=============================
+Culling idle notebook servers
+=============================
+
+The idle culler is a hub-managed service that automatically shuts down idle
+single-user notebook servers in order to free up resources. After culling, any
+in-memory data will be lost.
+
+
+Disabling the idle culler
+=========================
+
+The idle culling service is enabled by default. To disable it, use the following
+command:
+
+.. code-block:: bash
+
+   sudo tljh-config set services.cull.enabled False
+
+
+Configuring the idle culler
+===========================
+
+By **default**,  JupyterHub will:
+	* Run the culling process every minute.
+	* Cull any user servers that have been inactive for more than 10 minutes.
+
+The configuration options available are:
+
+Idle timeout
+------------
+
+The idle timeout (in seconds) can be configured using:
+
+.. code-block:: bash
+
+	sudo tljh-config set services.cull.timeout <max-idle-sec-before-server-is-culled>
+
+*By default services.cull.timeout = 600*
+
+Idle check interval
+-------------------
+
+The interval (in seconds) for checking for idle servers to cull can be configured using:
+
+.. code-block:: bash
+
+ 	sudo tljh-config set services.cull.every <number-of-sec-this-check-is-done>
+
+*By default services.cull.every = 60*
+
+Maximum age
+-----------
+
+The maximum age (in seconds) of servers that should be culled even if they are active
+can be configured using:
+
+.. code-block:: bash
+
+ 	sudo tljh-config set services.cull.max_age <server-max-age>
+
+*By default services.cull.max_age = 0*
+
+User culling
+------------
+
+In addition to servers, the users will also be culled if the following command is used:
+
+.. code-block:: bash
+
+ 	sudo tljh-config set services.cull.users True
+
+*By default services.cull.users = False*
+
+Concurrency
+-----------
+
+The number of concurrent requests made to the Hub ca be configured using:
+
+.. code-block:: bash
+
+ 	sudo tljh-config set services.cull.concurrency <number-of-concurrent-hub-requests>
+
+*By default services.cull.concurrency = 5*

integration-tests/test_hub.py

@@ -1,6 +1,7 @@
 import requests
 from hubtraf.user import User
 from hubtraf.auth.dummy import login_dummy
+from jupyterhub.utils import exponential_backoff
 import secrets
 import pytest
 from functools import partial
@@ -137,4 +138,99 @@ async def test_long_username():
             '-u', 'jupyterhub',
             '--no-pager'
         ])
-        raise
+        raise
+
+
+@pytest.mark.asyncio
+async def test_idle_server_culled():
+    """
+    User logs in, starts a server & stays idle for 1 min.
+    (the user's server should be culled during this period)
+    """
+    # This *must* be localhost, not an IP
+    # aiohttp throws away cookies if we are connecting to an IP!
+    hub_url = 'http://localhost'
+    username = secrets.token_hex(8)
+
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'auth.type', 'dummyauthenticator.DummyAuthenticator')).wait()
+    # Check every 10s for idle servers to cull
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.every', "10")).wait()
+    # Apart from servers, also cull users
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.users', "True")).wait()
+    # Cull servers and users after 60s of activity
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.max_age', "60")).wait()
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'reload')).wait()
+
+    async with User(username, hub_url, partial(login_dummy, password='')) as u:
+            await u.login()
+            # Start user's server
+            await u.ensure_server()
+            # Assert that the user exists
+            assert pwd.getpwnam(f'jupyter-{username}') is not None
+
+            # Check that we can get to the user's server
+            r = await u.session.get(u.hub_url / 'hub/api/users' / username,
+                headers={'Referer': str(u.hub_url / 'hub/')})
+            assert r.status == 200
+
+            async def _check_culling_done():
+                # Check that after 60s, the user and server have been culled and are not reacheable anymore
+                r = await u.session.get(u.hub_url / 'hub/api/users' / username,
+                    headers={'Referer': str(u.hub_url / 'hub/')})
+                print(r.status)
+                return r.status == 403
+
+            await exponential_backoff(
+                _check_culling_done,
+                "Server culling failed!",
+                timeout=100,
+            )
+
+@pytest.mark.asyncio
+async def test_active_server_not_culled():
+    """
+    User logs in, starts a server & stays idle for 30s
+    (the user's server should not be culled during this period).
+    """
+    # This *must* be localhost, not an IP
+    # aiohttp throws away cookies if we are connecting to an IP!
+    hub_url = 'http://localhost'
+    username = secrets.token_hex(8)
+
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'auth.type', 'dummyauthenticator.DummyAuthenticator')).wait()
+    # Check every 10s for idle servers to cull
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.every', "10")).wait()
+    # Apart from servers, also cull users
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.users', "True")).wait()
+    # Cull servers and users after 60s of activity
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'set', 'services.cull.max_age', "60")).wait()
+    assert 0 == await (await asyncio.create_subprocess_exec(*TLJH_CONFIG_PATH, 'reload')).wait()
+
+    async with User(username, hub_url, partial(login_dummy, password='')) as u:
+            await u.login()
+            # Start user's server
+            await u.ensure_server()
+            # Assert that the user exists
+            assert pwd.getpwnam(f'jupyter-{username}') is not None
+
+            # Check that we can get to the user's server
+            r = await u.session.get(u.hub_url / 'hub/api/users' / username,
+                headers={'Referer': str(u.hub_url / 'hub/')})
+            assert r.status == 200
+
+            async def _check_culling_done():
+                # Check that after 30s, we can still reach the user's server
+                r = await u.session.get(u.hub_url / 'hub/api/users' / username,
+                    headers={'Referer': str(u.hub_url / 'hub/')})
+                print(r.status)
+                return r.status != 200
+
+            try:
+                await exponential_backoff(
+                    _check_culling_done,
+                    "User's server is still reacheable!",
+                    timeout=30,
+                )
+            except TimeoutError:
+                # During the 30s timeout the user's server wasn't culled, which is what we intended.
+                pass

tests/test_configurer.py

@@ -3,6 +3,7 @@
 """
 
 import os
+import sys
 
 from tljh import configurer
 
@@ -187,6 +188,49 @@ def test_set_traefik_api():
     assert c.TraefikTomlProxy.traefik_api_password == '1234'
 
 
+def test_cull_service_default():
+    """
+    Test default cull service settings with no overrides
+    """
+    c = apply_mock_config({})
+
+    cull_cmd = [
+       sys.executable, '/srv/src/tljh/cull_idle_servers.py',
+       '--timeout=600', '--cull-every=60', '--concurrency=5',
+       '--max-age=0'
+    ]
+    assert c.JupyterHub.services == [{
+        'name': 'cull-idle',
+        'admin': True,
+        'command': cull_cmd,
+    }]
+
+
+def test_set_cull_service():
+    """
+    Test setting cull service options
+    """
+    c = apply_mock_config({
+        'services': {
+            'cull': {
+                'every': 10,
+                'users': True,
+                'max_age': 60
+            }
+        }
+    })
+    cull_cmd = [
+       sys.executable, '/srv/src/tljh/cull_idle_servers.py',
+       '--timeout=600', '--cull-every=10', '--concurrency=5',
+       '--max-age=60', '--cull-users'
+    ]
+    assert c.JupyterHub.services == [{
+        'name': 'cull-idle',
+        'admin': True,
+        'command': cull_cmd,
+    }]
+
+
 def test_load_secrets(tljh_dir):
     """
     Test loading secret files

tljh/configurer.py

@@ -9,6 +9,7 @@
 """
 
 import os
+import sys
 
 from .config import CONFIG_FILE, STATE_DIR
 from .yaml import yaml
@@ -55,6 +56,16 @@
     'user_environment': {
         'default_app': 'classic',
     },
+    'services': {
+        'cull': {
+            'enabled': True,
+            'timeout': 600,
+            'every': 60,
+            'concurrency': 5,
+            'users': False,
+            'max_age': 0
+        }
+    }
 }
 
 def load_config(config_file=CONFIG_FILE):
@@ -86,6 +97,7 @@ def apply_config(config_overrides, c):
     update_user_environment(c, tljh_config)
     update_user_account_config(c, tljh_config)
     update_traefik_api(c, tljh_config)
+    update_services(c, tljh_config)
 
 
 def set_if_not_none(parent, key, value):
@@ -191,6 +203,38 @@ def update_traefik_api(c, config):
     c.TraefikTomlProxy.traefik_api_password = config['traefik_api']['password']
 
 
+def set_cull_idle_service(config):
+    """
+    Set Idle Culler service
+    """
+    cull_cmd = [
+       sys.executable, '/srv/src/tljh/cull_idle_servers.py'
+    ]
+    cull_config = config['services']['cull']
+    print()
+
+    cull_cmd += ['--timeout=%d' % cull_config['timeout']]
+    cull_cmd += ['--cull-every=%d' % cull_config['every']]
+    cull_cmd += ['--concurrency=%d' % cull_config['concurrency']]
+    cull_cmd += ['--max-age=%d' % cull_config['max_age']]
+    if cull_config['users']:
+        cull_cmd += ['--cull-users']
+
+    cull_service = {
+        'name': 'cull-idle',
+        'admin': True,
+        'command': cull_cmd,
+    }
+
+    return cull_service
+
+
+def update_services(c, config):
+    c.JupyterHub.services = []
+    if config['services']['cull']['enabled']:
+        c.JupyterHub.services.append(set_cull_idle_service(config))
+
+
 def _merge_dictionaries(a, b, path=None, update=True):
     """
     Merge two dictionaries recursively.