The default AWS rate limit is 2 instances/sec, but in practice, we can achieve something like 0.6 instances/sec with the current code. That's because the create instance REST API call itself takes more than a second to return. To achieve even the default AWS rate (much less a potentially faster one which may be obtainable via support request), we need to alter the approach. This change does the following: * Paralellizes create API calls. We create a threadpool with (typically) 8 workers to execute create instance calls in the background. 2 or 3 workers should be sufficient to meet the 2/sec rate, more allows for the occasional longer execution time as well as a customized higher rate. We max out at 8 to protect nodepool from too many threads. * The state machine uses the new background create calls instead of synchronously creating instances. This allows other state machines to progress further (ie, advance to ssh keyscan faster in the case of a rush of requests). * Delete calls are batched. They don't take as long as create calls, yet their existence at all uses up rate limiting slots which could be used for creating instances. By batching deletes, we make more room for creates. * A bug in the RateLimiter could cause it not to record the initial time and therefore avoid actually rate limiting. This is fixed. * The RateLimiter is now thread-safe. * The default rate limit for AWS is changed to 2 requests/sec. * Documentation for the 'rate' parameter for the AWS driver is added. * Documentation for the 'rate' parameter for the Azure driver is corrected to describe the rate as requests/sec instead of delay between requests. Change-Id: Ida2cbc59928e183eb7da275ff26d152eae784cfe
18 KiB
zuul
AWS Driver
If using the AWS driver to upload diskimages, see VM Import/Export service role for information on configuring the required permissions in AWS. You must also create an S3 Bucket for use by Nodepool.
Selecting the aws
driver adds the following options to
the providers
section
of the configuration.
providers.[aws]
An AWS provider's resources are partitioned into groups called pool (see providers.[aws].pools
for details), and within a
pool, the node types which are to be made available are listed (see
providers.[aws].pools.labels
for details).
See Boto Configuration for information on how to configure credentials and other settings for AWS access in Nodepool's runtime environment.
Note
For documentation purposes the option names are prefixed
providers.[aws]
to disambiguate from other drivers, but
[aws]
is not required in the configuration (e.g. below
providers.[aws].pools
refers to the pools
key
in the providers
section when the aws
driver
is selected).
Example:
providers:
- name: ec2-us-west-2
driver: aws
region-name: us-west-2
cloud-images:
- name: debian9
image-id: ami-09c308526d9534717
username: admin
pools:
- name: main
max-servers: 5
subnet-id: subnet-0123456789abcdef0
security-group-id: sg-01234567890abcdef
labels:
- name: debian9
cloud-image: debian9
instance-type: t3.medium
iam-instance-profile:
arn: arn:aws:iam::123456789012:instance-profile/s3-read-only
key-name: zuul
tags:
key1: value1
- name: debian9-large
cloud-image: debian9
instance-type: t3.large
key-name: zuul
tags:
key1: value1
key2: value2
name
A unique name for this provider configuration.
region-name
Name of the AWS region to interact with.
profile-name
The AWS credentials profile to load for this provider. If unspecified the boto3 library will select a profile.
See Boto Configuration for more information.
rate
The number of operations per second to perform against the provider.
boot-timeout
Once an instance is active, how long to try connecting to the image via SSH. If the timeout is exceeded, the node launch is aborted and the instance deleted.
launch-timeout
The time to wait from issuing the command to create a new instance until that instance is reported as "active". If the timeout is exceeded, the node launch is aborted and the instance deleted.
launch-retries
The number of times to retry launching a node before considering the request failed.
post-upload-hook
Filename of an optional script that can be called after an image has been uploaded to a provider but before it is taken into use. This is useful to perform last minute validation tests before an image is really used for build nodes. The script will be called as follows:
<SCRIPT> <PROVIDER> <EXTERNAL_IMAGE_ID> <LOCAL_IMAGE_FILENAME>
If the script returns with result code 0 it is treated as successful otherwise it is treated as failed and the image gets deleted.
object-storage
This section is only required when using Nodepool to upload diskimages.
bucket-name
The name of a bucket to use for temporary storage of diskimages while creating snapshots. The bucket must already exist.
image-format
The image format that should be requested from diskimage-builder and
also specified to AWS when importing images. One of: ova
,
vhd
, vhdx
, vmdk
, raw
(not all of which are supported by diskimage-builder).
cloud-images
Each entry in this section must refer to an entry in the labels
section.
cloud-images:
- name: ubuntu1804
image-id: ami-082fd9a18128c9e8c
username: ubuntu
- name: ubuntu1804-by-filters
image-filters:
- name: name
values:
- named-ami
username: ubuntu
- name: my-custom-win2k3
connection-type: winrm
username: admin
Each entry is a dictionary with the following keys
name
Identifier to refer this cloud-image from providers.[aws].pools.labels
section. Since this name appears elsewhere in the nodepool configuration
file, you may want to use your own descriptive name here and use
image-id
to specify the cloud image so that if the image id
changes on the cloud, the impact to your Nodepool configuration will be
minimal. However, if image-id
is not provided, this is
assumed to be the image id in the cloud.
image-id
If this is provided, it is used to select the image from the cloud
provider by ID. Either this field or providers.[aws].cloud-images.image-filters
must be
provided.
image-filters
If provided, this is used to select an AMI by filters. If the filters
provided match more than one image, the most recent will be returned.
Either this field or providers.[aws].cloud-images.image-id
must be
provided.
Each entry is a dictionary with the following keys
name
The filter name. See Boto describe images for a list of valid filters.
values
A list of string values on which to filter.
username
The username that a consumer should use when connecting to the node.
python-path
The path of the default python interpreter. Used by Zuul to set
ansible_python_interpreter
. The special value
auto
will direct Zuul to use inbuilt Ansible logic to
select the interpreter on Ansible >=2.8, and default to
/usr/bin/python2
for earlier versions.
connection-type
The connection type that a consumer should use when connecting to the node. For most images this is not necessary. However when creating Windows images this could be 'winrm' to enable access via ansible.
connection-port
The port that a consumer should use when connecting to the node. For most diskimages this is not necessary. This defaults to 22 for ssh and 5986 for winrm.
shell-type
The shell type of the node's default shell executable. Used by Zuul
to set ansible_shell_type
. This setting should only be
used
- For a windows image with the experimental connection-type
ssh
in which casecmd
orpowershell
should be set and reflect the node'sDefaultShell
configuration. - If the default shell is not Bourne compatible (sh), but instead e.g.
csh
orfish
, and the user is aware that there is a long-standing issue withansible_shell_type
in combination withbecome
.
diskimages
Each entry in a provider's diskimages
section must correspond to an entry in diskimages
. Such an entry indicates that the
corresponding diskimage should be uploaded for use in this provider.
Additionally, any nodes that are created using the uploaded image will
have the associated attributes (such as flavor or metadata).
If an image is removed from this section, any previously uploaded images will be deleted from the provider.
diskimages:
- name: bionic
pause: False
- name: windows
connection-type: winrm
connection-port: 5986
Each entry is a dictionary with the following keys
name
Identifier to refer this image from providers.[aws].pools.labels
and diskimages
sections.
pause
When set to True, nodepool-builder will not upload the image to the provider.
username
The username that should be used when connecting to the node.
connection-type
The connection type that a consumer should use when connecting to the
node. For most diskimages this is not necessary. However when creating
Windows images this could be winrm
to enable access via
ansible.
connection-port
The port that a consumer should use when connecting to the node. For most diskimages this is not necessary. This defaults to 22 for ssh and 5986 for winrm.
python-path
The path of the default python interpreter. Used by Zuul to set
ansible_python_interpreter
. The special value
auto
will direct Zuul to use inbuilt Ansible logic to
select the interpreter on Ansible >=2.8, and default to
/usr/bin/python2
for earlier versions.
shell-type
The shell type of the node's default shell executable. Used by Zuul
to set ansible_shell_type
. This setting should only be
used
- For a windows image with the experimental connection-type
ssh
in which casecmd
orpowershell
should be set and reflect the node'sDefaultShell
configuration. - If the default shell is not Bourne compatible (sh), but instead e.g.
csh
orfish
, and the user is aware that there is a long-standing issue withansible_shell_type
in combination withbecome
.
pools
A pool defines a group of resources from an AWS provider. Each pool has a maximum number of nodes which can be launched from it, along with a number of cloud-related attributes used when launching nodes.
name
A unique name within the provider for this pool of resources.
priority
The priority of this provider pool (a lesser number is a higher priority). Nodepool launchers will yield requests to other provider pools with a higher priority as long as they are not paused. This means that in general, higher priority pools will reach quota first before lower priority pools begin to be used.
This setting may be specified at the provider level in order to apply to all pools within that provider, or it can be overridden here for a specific pool.
node-attributes
A dictionary of key-value pairs that will be stored with the node data in ZooKeeper. The keys and values can be any arbitrary string.
subnet-id
If provided, specifies the subnet to assign to the primary network interface of nodes.
security-group-id
If provided, specifies the security group ID to assign to the primary network interface of nodes.
public-ip-address
Deprecated alias for providers.[aws].pools.public-ipv4
.
public-ipv4
Specify if a public IPv4 address shall be attached to nodes.
public-ipv6
Specify if a public IPv6 address shall be attached to nodes.
use-internal-ip
If a public IP is attached but Nodepool should prefer the private IP, set this to true.
host-key-checking
Whether to validate SSH host keys. When true, this helps ensure that nodes are ready to receive SSH connections before they are supplied to the requestor. When set to false, nodepool-launcher will not attempt to ssh-keyscan nodes after they are booted. Disable this if nodepool-launcher and the nodes it launches are on different networks, where the launcher is unable to reach the nodes directly, or when using Nodepool with non-SSH node platforms. The default value is true.
labels
Each entry in a pool's labels section indicates that the corresponding label is available for use in this pool. When creating nodes for a label, the flavor-related attributes in that label's section will be used.
labels:
- name: bionic
instance-type: m5a.large
Each entry is a dictionary with the following keys
name
Identifier to refer to this label.
cloud-image
Refers to the name of an externally managed image in the cloud that already exists on the provider. The value of
cloud-image
should match thename
of a previously configured entry from thecloud-images
section of the provider. Seeproviders.[aws].cloud-images
. Mutually exclusive withproviders.[aws].pools.labels.diskimage
diskimage
Refers to provider's diskimages, see
providers.[aws].diskimages
. Mutually exclusive withproviders.[aws].pools.labels.cloud-image
ebs-optimized
Indicates whether EBS optimization (additional, dedicated throughput between Amazon EC2 and Amazon EBS,) has been enabled for the instance.
instance-type
Name of the flavor to use.
iam-instance-profile
Used to attach an iam instance profile. Useful for giving access to services without needing any secrets.
name
Name of the instance profile. Mutually exclusive with
providers.[aws].pools.labels.iam-instance-profile.arn
arn
ARN identifier of the profile. Mutually exclusive with
providers.[aws].pools.labels.iam-instance-profile.name
key-name
The name of a keypair that will be used when booting each server.
volume-type
If given, the root EBS volume type
volume-size
If given, the size of the root EBS volume, in GiB.
userdata
A string of userdata for a node. Example usage is to install cloud-init package on image which will apply the userdata. Additional info about options in cloud-config: https://cloudinit.readthedocs.io/en/latest/topics/examples.html
tags
A dictionary of tags to add to the EC2 instances. Values must be supplied as strings.