3 navsection: installguide
4 title: Install Node Manager
7 Arvados Node Manager provides elastic computing for Arvados and SLURM by creating and destroying virtual machines on demand. Node Manager currently supports Amazon Web Services (AWS), Google Cloud Platform (GCP) and Microsoft Azure.
9 Note: node manager is only required for elastic computing cloud environments. Fixed size clusters do not require node manager.
13 Node manager may run anywhere, however it must be able to communicate with the cloud provider's APIs, and use the command line tools @sinfo@, @squeue@ and @scontrol@ to communicate with the cluster's SLURM controller.
15 On Debian-based systems:
18 <pre><code>~$ <span class="userinput">sudo apt-get install arvados-node-manager</span>
22 On Red Hat-based systems:
25 <pre><code>~$ <span class="userinput">sudo yum install arvados-node-manager</span>
29 h2. Create compute image
31 Configure a virtual machine following the "instructions to set up a compute node.":{{site.baseurl}}/install/crunch2-slurm/install-compute-node.html and set it up to run a "ping script":{{site.baseurl}}/install/install-compute-ping.html at boot.
33 Create a virtual machine image using the commands provided by your cloud provider. We recommend using a tool such as "Packer":https://www.packer.io/ to automate this process.
35 Configure node manager to use the image with the @image@ or @image_id@ parameter.
37 h2. Configure node manager
39 The configuration file at @/etc/arvados-node-manager/config.ini@ . Some configuration details are specific to the cloud provider you are using:
41 * "Amazon Web Services":#aws
42 * "Google Cloud Platform":#gcp
43 * "Microsoft Azure":#azure
45 h3(#aws). Amazon Web Services
48 # EC2 configuration for Arvados Node Manager.
49 # All times are in seconds unless specified otherwise.
52 # The management server responds to http://addr:port/status.json with
53 # a snapshot of internal state.
55 # Management server listening address (default 127.0.0.1)
58 # Management server port number (default -1, server is disabled)
62 # The dispatcher can customize the start and stop procedure for
63 # cloud nodes. For example, the SLURM dispatcher drains nodes
64 # through SLURM before shutting them down.
67 # Node Manager will ensure that there are at least this many nodes running at
68 # all times. If node manager needs to start new idle nodes for the purpose of
69 # satisfying min_nodes, it will use the cheapest node type. However, depending
70 # on usage patterns, it may also satisfy min_nodes by keeping alive some
71 # more-expensive nodes
74 # Node Manager will not start any compute nodes when at least this
78 # Upper limit on rate of spending (in $/hr), will not boot additional nodes
79 # if total price of already running nodes meets or exceeds this threshold.
80 # default 0 means no limit.
83 # Poll EC2 nodes and Arvados for new information every N seconds.
86 # Polls have exponential backoff when services fail to respond.
87 # This is the longest time to wait between polls.
90 # If Node Manager can't succesfully poll a service for this long,
91 # it will never start or stop compute nodes, on the assumption that its
92 # information is too outdated.
93 poll_stale_after = 600
95 # If Node Manager boots a cloud node, and it does not pair with an Arvados
96 # node before this long, assume that there was a cloud bootstrap failure and
97 # shut it down. Note that normal shutdown windows apply (see the Cloud
98 # section), so this should be shorter than the first shutdown window value.
99 boot_fail_after = 1800
101 # "Node stale time" affects two related behaviors.
102 # 1. If a compute node has been running for at least this long, but it
103 # isn't paired with an Arvados node, do not shut it down, but leave it alone.
104 # This prevents the node manager from shutting down a node that might
105 # actually be doing work, but is having temporary trouble contacting the
107 # 2. When the Node Manager starts a new compute node, it will try to reuse
108 # an Arvados node that hasn't been updated for this long.
109 node_stale_after = 14400
111 # Scaling factor to be applied to nodes' available RAM size. Usually there's a
112 # variable discrepancy between the advertised RAM value on cloud nodes and the
113 # actual amount available.
114 # If not set, this value will be set to 0.95
115 node_mem_scaling = 0.95
117 # File path for Certificate Authorities
118 certs_file = /etc/ssl/certs/ca-certificates.crt
122 file = /var/log/arvados/node-manager.log
124 # Log level for most Node Manager messages.
125 # Choose one of DEBUG, INFO, WARNING, ERROR, or CRITICAL.
126 # WARNING lets you know when polling a service fails.
127 # INFO additionally lets you know when a compute node is started or stopped.
130 # You can also set different log levels for specific libraries.
131 # Pykka is the Node Manager's actor library.
132 # Setting this to DEBUG will display tracebacks for uncaught
133 # exceptions in the actors, but it's also very chatty.
136 # Setting apiclient to INFO will log the URL of every Arvados API request.
140 host = zyxwv.arvadosapi.com
141 token = ARVADOS_TOKEN
144 # Accept an untrusted SSL certificate from the API server?
150 # It's usually most cost-effective to shut down compute nodes during narrow
151 # windows of time. For example, EC2 bills each node by the hour, so the best
152 # time to shut down a node is right before a new hour of uptime starts.
153 # Shutdown windows define these periods of time. These are windows in
154 # full minutes, separated by commas. Counting from the time the node is
155 # booted, the node WILL NOT shut down for N1 minutes; then it MAY shut down
156 # for N2 minutes; then it WILL NOT shut down for N3 minutes; and so on.
157 # For example, "54, 5, 1" means the node may shut down from the 54th to the
158 # 59th minute of each hour of uptime.
159 # Specify at least two windows. You can add as many as you need beyond that.
160 shutdown_windows = 54, 5, 1
169 # This section defines filters that find compute nodes.
170 # Tags that you specify here will automatically be added to nodes you create.
171 # Replace colons in Amazon filters with underscores
172 # (e.g., write "tag:mytag" as "tag_mytag").
173 instance-state-name = running
174 tag_arvados-class = dynamic-compute
178 # New compute nodes will send pings to Arvados at this host.
179 # You may specify a port, and use brackets to disambiguate IPv6 addresses.
180 ping_host = hostname:port
182 # Give the name of an SSH key on AWS...
185 # ... or a file path for an SSH key that can log in to the compute node.
186 # (One or the other, not both.)
189 # The EC2 IDs of the image and subnet compute nodes should use.
193 # Comma-separated EC2 IDs for the security group(s) assigned to each
195 security_groups = idstring1, idstring2
198 # You can define any number of Size sections to list EC2 sizes you're
199 # willing to use. The Node Manager should boot the cheapest size(s) that
200 # can run jobs in the queue.
202 # Each size section MUST define the number of cores are available in this
203 # size class (since libcloud does not provide any consistent API for exposing
205 # You may also want to define the amount of scratch space (expressed
206 # in GB) for Crunch jobs. You can also override Amazon's provided
207 # data fields (such as price per hour) by setting them here.
220 h3(#gcp). Google Cloud Platform
223 # Google Compute Engine configuration for Arvados Node Manager.
224 # All times are in seconds unless specified otherwise.
227 # The management server responds to http://addr:port/status.json with
228 # a snapshot of internal state.
230 # Management server listening address (default 127.0.0.1)
233 # Management server port number (default -1, server is disabled)
237 # Node Manager will ensure that there are at least this many nodes running at
238 # all times. If node manager needs to start new idle nodes for the purpose of
239 # satisfying min_nodes, it will use the cheapest node type. However, depending
240 # on usage patterns, it may also satisfy min_nodes by keeping alive some
241 # more-expensive nodes
244 # Node Manager will not start any compute nodes when at least this
245 # running at all times. By default, these will be the cheapest node size.
248 # Poll compute nodes and Arvados for new information every N seconds.
251 # Upper limit on rate of spending (in $/hr), will not boot additional nodes
252 # if total price of already running nodes meets or exceeds this threshold.
253 # default 0 means no limit.
256 # Polls have exponential backoff when services fail to respond.
257 # This is the longest time to wait between polls.
260 # If Node Manager can't succesfully poll a service for this long,
261 # it will never start or stop compute nodes, on the assumption that its
262 # information is too outdated.
263 poll_stale_after = 600
265 # "Node stale time" affects two related behaviors.
266 # 1. If a compute node has been running for at least this long, but it
267 # isn't paired with an Arvados node, do not shut it down, but leave it alone.
268 # This prevents the node manager from shutting down a node that might
269 # actually be doing work, but is having temporary trouble contacting the
271 # 2. When the Node Manager starts a new compute node, it will try to reuse
272 # an Arvados node that hasn't been updated for this long.
273 node_stale_after = 14400
275 # Scaling factor to be applied to nodes' available RAM size. Usually there's a
276 # variable discrepancy between the advertised RAM value on cloud nodes and the
277 # actual amount available.
278 # If not set, this value will be set to 0.95
279 node_mem_scaling = 0.95
281 # File path for Certificate Authorities
282 certs_file = /etc/ssl/certs/ca-certificates.crt
286 file = /var/log/arvados/node-manager.log
288 # Log level for most Node Manager messages.
289 # Choose one of DEBUG, INFO, WARNING, ERROR, or CRITICAL.
290 # WARNING lets you know when polling a service fails.
291 # INFO additionally lets you know when a compute node is started or stopped.
294 # You can also set different log levels for specific libraries.
295 # Pykka is the Node Manager's actor library.
296 # Setting this to DEBUG will display tracebacks for uncaught
297 # exceptions in the actors, but it's also very chatty.
300 # Setting apiclient to INFO will log the URL of every Arvados API request.
304 host = zyxwv.arvadosapi.com
305 token = ARVADOS_TOKEN
308 # Accept an untrusted SSL certificate from the API server?
314 # Shutdown windows define periods of time when a node may and may not
315 # be shut down. These are windows in full minutes, separated by
316 # commas. Counting from the time the node is booted, the node WILL
317 # NOT shut down for N1 minutes; then it MAY shut down for N2 minutes;
318 # then it WILL NOT shut down for N3 minutes; and so on. For example,
319 # "54, 5, 1" means the node may shut down from the 54th to the 59th
320 # minute of each hour of uptime.
321 # GCE bills by the minute, and does not provide information about when
322 # a node booted. Node Manager will store this information in metadata
323 # when it boots a node; if that information is not available, it will
324 # assume the node booted at the epoch. These shutdown settings are
325 # very aggressive. You may want to adjust this if you want more
326 # continuity of service from a single node.
327 shutdown_windows = 20, 999999
330 user_id = client_email_address@developer.gserviceaccount.com
331 key = path_to_certificate.pem
332 project = project-id-from-google-cloud-dashboard
335 # Valid location (zone) names: https://cloud.google.com/compute/docs/zones
336 datacenter = us-central1-a
338 # Optional settings. For full documentation see
339 # http://libcloud.readthedocs.org/en/latest/compute/drivers/gce.html#libcloud.compute.drivers.gce.GCENodeDriver
341 # auth_type = SA # SA, IA or GCE
342 # scopes = https://www.googleapis.com/auth/compute
346 # A comma-separated list of tags that must be applied to a node for it to
347 # be considered a compute node.
348 # The driver will automatically apply these tags to nodes it creates.
349 tags = zyxwv, compute
352 # New compute nodes will send pings to Arvados at this host.
353 # You may specify a port, and use brackets to disambiguate IPv6 addresses.
354 ping_host = hostname:port
356 # A file path for an SSH key that can log in to the compute node.
359 # The GCE image name and network zone name to use when creating new nodes.
361 # network = your_network_name
363 # JSON string of service account authorizations for this cluster.
364 # See http://libcloud.readthedocs.org/en/latest/compute/drivers/gce.html#specifying-service-account-scopes
365 # service_accounts = [{'email':'account@example.com', 'scopes':['storage-ro']}]
368 # You can define any number of Size sections to list node sizes you're
369 # willing to use. The Node Manager should boot the cheapest size(s) that
370 # can run jobs in the queue.
372 # The Size fields are interpreted the same way as with a libcloud NodeSize:
373 # http://libcloud.readthedocs.org/en/latest/compute/api.html#libcloud.compute.base.NodeSize
375 # See https://cloud.google.com/compute/docs/machine-types for a list
376 # of known machine types that may be used as a Size parameter.
378 # Each size section MUST define the number of cores are available in this
379 # size class (since libcloud does not provide any consistent API for exposing
381 # You may also want to define the amount of scratch space (expressed
382 # in GB) for Crunch jobs.
383 # You can also override Google's provided data fields (such as price per hour)
384 # by setting them here.
397 h3(#azure). Microsoft Azure
400 # Azure configuration for Arvados Node Manager.
401 # All times are in seconds unless specified otherwise.
404 # The management server responds to http://addr:port/status.json with
405 # a snapshot of internal state.
407 # Management server listening address (default 127.0.0.1)
410 # Management server port number (default -1, server is disabled)
414 # The dispatcher can customize the start and stop procedure for
415 # cloud nodes. For example, the SLURM dispatcher drains nodes
416 # through SLURM before shutting them down.
419 # Node Manager will ensure that there are at least this many nodes running at
420 # all times. If node manager needs to start new idle nodes for the purpose of
421 # satisfying min_nodes, it will use the cheapest node type. However, depending
422 # on usage patterns, it may also satisfy min_nodes by keeping alive some
423 # more-expensive nodes
426 # Node Manager will not start any compute nodes when at least this
430 # Upper limit on rate of spending (in $/hr), will not boot additional nodes
431 # if total price of already running nodes meets or exceeds this threshold.
432 # default 0 means no limit.
435 # Poll Azure nodes and Arvados for new information every N seconds.
438 # Polls have exponential backoff when services fail to respond.
439 # This is the longest time to wait between polls.
442 # If Node Manager can't succesfully poll a service for this long,
443 # it will never start or stop compute nodes, on the assumption that its
444 # information is too outdated.
445 poll_stale_after = 600
447 # If Node Manager boots a cloud node, and it does not pair with an Arvados
448 # node before this long, assume that there was a cloud bootstrap failure and
449 # shut it down. Note that normal shutdown windows apply (see the Cloud
450 # section), so this should be shorter than the first shutdown window value.
451 boot_fail_after = 1800
453 # "Node stale time" affects two related behaviors.
454 # 1. If a compute node has been running for at least this long, but it
455 # isn't paired with an Arvados node, do not shut it down, but leave it alone.
456 # This prevents the node manager from shutting down a node that might
457 # actually be doing work, but is having temporary trouble contacting the
459 # 2. When the Node Manager starts a new compute node, it will try to reuse
460 # an Arvados node that hasn't been updated for this long.
461 node_stale_after = 14400
463 # Scaling factor to be applied to nodes' available RAM size. Usually there's a
464 # variable discrepancy between the advertised RAM value on cloud nodes and the
465 # actual amount available.
466 # If not set, this value will be set to 0.95
467 node_mem_scaling = 0.95
469 # File path for Certificate Authorities
470 certs_file = /etc/ssl/certs/ca-certificates.crt
474 file = /var/log/arvados/node-manager.log
476 # Log level for most Node Manager messages.
477 # Choose one of DEBUG, INFO, WARNING, ERROR, or CRITICAL.
478 # WARNING lets you know when polling a service fails.
479 # INFO additionally lets you know when a compute node is started or stopped.
482 # You can also set different log levels for specific libraries.
483 # Pykka is the Node Manager's actor library.
484 # Setting this to DEBUG will display tracebacks for uncaught
485 # exceptions in the actors, but it's also very chatty.
488 # Setting apiclient to INFO will log the URL of every Arvados API request.
492 host = zyxwv.arvadosapi.com
493 token = ARVADOS_TOKEN
496 # Accept an untrusted SSL certificate from the API server?
502 # Shutdown windows define periods of time when a node may and may not be shut
503 # down. These are windows in full minutes, separated by commas. Counting from
504 # the time the node is booted, the node WILL NOT shut down for N1 minutes; then
505 # it MAY shut down for N2 minutes; then it WILL NOT shut down for N3 minutes;
506 # and so on. For example, "20, 999999" means the node may shut down between
507 # the 20th and 999999th minutes of uptime.
508 # Azure bills by the minute, so it makes sense to agressively shut down idle
509 # nodes. Specify at least two windows. You can add as many as you need beyond
511 shutdown_windows = 20, 999999
514 # Use "azure account list" with the azure CLI to get these values.
515 tenant_id = 00000000-0000-0000-0000-000000000000
516 subscription_id = 00000000-0000-0000-0000-000000000000
518 # The following directions are based on
519 # https://azure.microsoft.com/en-us/documentation/articles/resource-group-authenticate-service-principal/
521 # azure config mode arm
522 # azure ad app create --name "<Your Application Display Name>" --home-page "<https://YourApplicationHomePage>" --identifier-uris "<https://YouApplicationUri>" --password <Your_Password>
523 # azure ad sp create "<Application_Id>"
524 # azure role assignment create --objectId "<Object_Id>" -o Owner -c /subscriptions/{subscriptionId}/
526 # Use <Application_Id> for "key" and the <Your_Password> for "secret"
528 key = 00000000-0000-0000-0000-000000000000
534 # The resource group in which the compute node virtual machines will be created
536 ex_resource_group = ArvadosResourceGroup
539 # The compute node image, as a link to a VHD in Azure blob store.
540 image = https://example.blob.core.windows.net/system/Microsoft.Compute/Images/images/zyxwv-compute-osDisk.vhd
542 # Path to a local ssh key file that will be used to provision new nodes.
543 ssh_key = /home/arvadosuser/.ssh/id_rsa.pub
545 # The account name for the admin user that will be provisioned on new nodes.
546 ex_user_name = arvadosuser
548 # The Azure storage account that will be used to store the node OS disk images.
549 ex_storage_account = arvadosstorage
551 # The virtual network the VMs will be associated with.
552 ex_network = ArvadosNetwork
554 # Optional subnet of the virtual network.
558 tag_arvados-class = dynamic-compute
561 # the API server to ping
562 ping_host = hostname:port
564 # You can define any number of Size sections to list Azure sizes you're willing
565 # to use. The Node Manager should boot the cheapest size(s) that can run jobs
566 # in the queue. You must also provide price per hour as the Azure driver
567 # compute currently does not report prices.
569 # See https://azure.microsoft.com/en-us/pricing/details/virtual-machines/
570 # for a list of known machine types that may be used as a Size parameter.
572 # Each size section MUST define the number of cores are available in this
573 # size class (since libcloud does not provide any consistent API for exposing
575 # You may also want to define the amount of scratch space (expressed
576 # in GB) for Crunch jobs. You can also override Microsoft's provided
577 # data fields by setting them here.
591 $ arvados-node-manager --config /etc/arvados-node-manager/config.ini