Schema definition
Schema Definition
Application Template Schema
This configuration template is the basis for all environments you will create for this application. Each of the sections and directives file help to create the configuration for your specific environment. The environment_templates section describes differences between your ephemeral and permanent templates and you select one of these when creating an environment. Each section and directive will be described in detail below in this document.
auto_deploy
If true, environments will deploy whenever you push to the corresponding repo and tracking branch.
context
This value is used by your application to deploy to a specific cluster. If you have your own EKS cluster through Release you can change this value to match that cluster/s, but if not use the generated value.
domain
The domain name where your applications will be hosted. These domains must be AWS Route 53 hosted domains. Release supports first and second level domains. (i.e. domain.com or release.domain.com)
execution_type
Determines whether the app creates Server or Runnable Environments
git_fetch_depth
Configures the fetch depth for Git operations in builds. Defaults to fetching the repository's complete Git history. Setting this to 1 will result in a shallow clone and can speed up builds for larger repositories. Defines s3 buckets that can be mounted as volumes in your services.
mode
Mode is a configuration directive that you can use (it is set as an environment variable in your containers) if useful. e.x 'development' or 'production' or 'test'
parallelize_app_imports
If there are no dependencies for the order in which in the apps deploy, use parallelize_app_imports to deploy all the apps at the same time.
tracking_branch
By default this will be the default branch of your repository, but it can be changed to any branch you would like to track with you environments.
tracking_tag
A specific git tag that you want your environments to track. You must unset tracking_branch if you use tracking_tag.
app_imports
App Imports are a way to connect multiple apps together. When you create an environment on one application, the apps that you import will also get environments created in the same namespace. Click here for more info.
builds
You can specify builds at the top level to be pulled in during the services sections. See the builds section for details.
cron_jobs
Cron Jobs are Jobs that run on a schedule. Cron jobs allow you to periodically execute commands within a namespace. They can be used for warming up caches, running database maintenance, etc. Click here for more info.
custom_links
Custom Links are defined as an array of key/value pairs, under the custom_links directive. The key is a name, and the value is any URL you want. The values can be hardcoded URLs or utilize variable substitution.
development_environment
This allows you to connect from a local machine to the remote environment and sync files and folders. Click here for more info.
environment_templates
These templates are used when creating an environment. They allow you to override or change any of the defaults in this file for particular type of environments: ephemeral or permanent. Click here for more info.
hostnames
Hostnames are defined as an array of key/value pairs, under the hostnames directive. The key is a service name, and the value is the hostname. These can be hardcoded hostnames or utilize variable substitution. They are auto-generated for any service with a node_port or static build.
infrastructure
Infrastructure runners are specialized jobs that execute infrastructure as code plans during deployment.
ingress
Ingress settings that can control the behavior and functionality of the NGINX ingress controller to access HTTP services in your cluster.
jobs
Jobs are like services except they run to completion. Examples include database migrations, asset compilation, etc. The inherit the image from a service, and run a command that ultimately terminates. Click here for more info.
node_selector
Node Selectors allow you to assign workloads to particular nodes based on common labels such as kubernetes.io/os=windows and kubernetes.io/arch=arm64. Click here for more information.
notifications
Allow you to define where and which events to be notified. Click here for more info.
parameters
These parameters allows you to collect info from the user at deploy time. You may interpolate these in your configuration and/or use them as inline environment variables for your service, jobs, etc. Click here for more info.
resources
Default resources for all of your services. The structure and values are based on https://kubernetes.io/docs/concepts/configuration/manage-resources-containers/. Click here for more info. For examples of why and how to override these defaults check out Managing Service Resources.
routes
Routes allow an easy way define multiple endpoints per service. Routes allow for edge routing rewrites, authentication and provide full support for ngnix ingress rules.
rules
Rules are based on https://kubernetes.github.io/ingress-nginx/user-guide/basic-usage/ and allow an easy way define multiple endpoints per service. They consist of three parts: service name, hostnames, and a path. Release will take this configuration and create an Nginx Ingress deployment to handle your routing automatically.
s3_volumes
Defines s3 buckets that can be mounted as volumes in your services
service_accounts
Allow you to define service accounts that can be used to control the cloud permissions assumed by your workloads (services, jobs, cron jobs, etc.)
services
These services define the most important part/s of your application. The can represent services Release builds from your repositories, off-the-shelf containers (postgres, redis, elasticsearch, etc), external services you need to connect to or even services from other applications you have that are needed in this application also. Click here for more info.
shared_volumes
Shared Volumes creates a PersistentVolumeClaim that is written to/read from by multiple services.
sidecars
Top level sidecar definitions allow you to create reusable containers that can be applied to several services defined with in your application. These are useful for log aggregation. Click here for more info.
workflows
Workflows are an ordered list of what must be done to deploy new configuration or code to your environments. They are a combination of services and jobs (if you have them). Click here for more info.
There are three workflows Release supports by default: setup, patch, and teardown. When a new environment is created setup is ran, and when code is pushed a patch is run against that environment. Whenever a an environment is destroyed the teardown workflow is run.
Release also supports user defined workflows that can be run in a one-off manner.
workspaces
Workspaces allow you to assemble multiple repositories or other data sources into a directory tree that can be shared by any number of containers in your environment. Click here for more info.
Hostnames or Rules
Hostnames or Rules can be both be used to define entry points to your services. They can not be used together at the same level in the config. In other words, you can't have default hostnames and rules, but you could have default hostnames and then use rules inside the environment_templates section of the file.
Hostnames
Hostnames are defined as an array of key/value pairs, under the hostnames directive. The key is a service name, and the value is the hostname. These can be hardcoded hostnames or utilize variable substitution. They are auto-generated for any service with a node_port or static build.
By default, Hostnames are generated using two variables env_id and domain. env_id is a randomly generated string for ephemeral environments or the name of the environment for permanent ones. Using some amount of random values allows Release to bring up any number of ephemeral environments on the same domain without conflicts. Domain is taken directly from your configuration file.
Rules
Rules are based on https://kubernetes.github.io/ingress-nginx/user-guide/basic-usage/ and allow an easy way define multiple endpoints per service. They consist of three parts: service name, hostnames, and a path. Release will take this configuration and create an Nginx Ingress deployment to handle your routing automatically. The visibility parameter will determine how the URL is reachable on the Internet (or privately via a VPC)
Rules Schema
Rules Example
App Imports
App Imports are optional and not present in the Application Template by default.
Example: App Imports excluding a service
Example: App Imports ignoring deployments
Example: App Imports with many apps utilizing the parallel deploys
You can optionally customize the order in which the current application is deployed with the special name $self. If not present, the current app is always deployed last.
Example: App Imports with custom ordering for the current application
You can optionally make certain app imports conditional upon whether or not a pull request has one of the specified labels.
Example: App Imports with pull request label filtering
Exclude Services
Allows the removal of duplicate services during App Imports
Builds
Top-level builds section for docker builds. Can be used for special docker images, especially jobs, init containers, and sidecars. If you are using build directives for a single service, please see the build section.
Example: A builds stanza for an init container. Please notice the use of
has_repoandnamefields in particular.
Cron Jobs
Cron Job containers allow you to define additional workloads run on a schedule. Cron Jobs can be used for many different tasks like database maintenance, reporting, warming caches by accessing other containers in the namespace, etc.
Each cron job entry has a mutually exclusive requirement where either image or from_services must be present.
Example cron job definitions to poll the frontend service and ping Redis
parallelism, completions, and concurrency_policy are ways to control how many pods will be spun up for jobs and how preemption will work. By default, a minimum of one job will run successfully to be considered passing. Also by default, we set concurrency_policy to equal Forbid rather than the default Kubernetes setting of Allow. We have found that the default of Allow creates problems for long running jobs or jobs that are intensive and need to be scheduled on a smaller cluster. For example, if a job runs for ten minutes but is scheduled every five minutes, then Kubernetes will gladly keep starting new jobs indefinitely because it does not think the job is finished. This can quickly overwhelm resources. You can use Forbid to prevent rescheduling jobs that should not be rescheduled even if they are not run or fail to start.
A few examples follow.
An example of the default settings (same as leaving them blank).
An example of a job that will run two polling jobs roughly simultaneously every ten minutes. Two jobs must succeed for the job to be marked complete; if it does not finish within 10 minutes, then the
Replacepolicy will kill the previous job and start a new one in its place.
An example of a queue-pulling job that will run 3 threads of self-synchronising pods and usually takes six runs to complete. The setting of
Allowwill ensure the job starts again if the scheduler decides the jobs did not finish or started late due to resource constraints on the cluster. Please note:completion_modeis not available until v1.24 is supported.
completions
Integer amount greater than zero to indicate how many successful runs should be considered finished. Usually you would set this value equal or greater than parallelism but it might be possible to set it less if you do not care about wasted pods being scheduled. Depending on parallelism and concurrency_policy, the combination of settings may cause jobs to be run multiple times in excess of this value. See the Kubernetes documentation
concurrency_policy
One of Allow, Forbid, or Replace. Kubernetes defaults to Allow which allows jobs to be rescheduled and started if they have failed or haven't started or haven't finished yet. We prefer to set Forbid because it prevents pods from being started or restarted again, which is much safer. The option to use Replace means that if a job is failed or stalled then the previous job start will be killed (if it is still running) before being started on a new pod.
from_services
A reference to the service name to use as the basis for executing the cron job. Parameters from the service will be copied into creating this cron job.
has_repo
Use an internal repository built by Release, or not.
image
A reference to the docker image to execute the job, use if from_services is not a good fit
name
What's in a name? That which we call a rose/By any other name would smell as sweet.
parallelism
Integer amount of number of pods that can run in parallel. Set to 0 to disable the cron job. See the Kubernetes documentation. This controls how many pods are potentially running at the same time during a scheduled run.
schedule
A string representing the schedule when a cron job will execute in the form of minute hour dayofmonth month dayofweek or @monthly, @weekly, etc. Read the Kubernetes docs
args
An array of arguments to be passed to the entrypoint of the container.
command
An array of arguments to be passed to override the entrypoint of the container.
Service Command
Pods running in Kubernetes typically have a default command that is run upon container start. The value specified in the command will override the supplied Docker ENTRYPOINT. Please note there is a lot of confusion for Kubernetes commands and Docker commands since they use similar and overlapping meanings for each. The command is specified as an EXECV array, not a string.
You can specify the override command that a shell will start with.
Development Environments
Development Environment allows you to configure an environment to be used for remote development. This allows you to connect from a local machine to the remote environment and sync files and folders.
Each service entry describes:
imageto use, if not using the same as the one defined on theservicecommandto run on the image, if not using the one defined on theservicesyncwhich files and folders to sync from a local machine to the remote containerport_forwardswhich ports to forward from the local machine to the remote container
Development Environment Example
Development Environment Services
Development Environment allows you to configure an environment to be used for remote development. This allows you to connect from a local machine to the remote environment and sync files and folders.
Port Forwards
Forward ports allows you to configure which local port(s) are mapped to the remote port(s) on your container.
Sync
Sync allows you to configure which files and folders are synchronized between a local machine and a remote container.
Environment Templates
There are two types of allowed and required templates: ephemeral and permanent. When creating a new environment, either manually or through a pull request one of these templates will be used to construct the configuration for that particular environment. If the template is empty you get the defaults contained in your Application Template, but these templates allow you to override any of the defaults.
The schema for these is a duplicate of the entire default configuration as it allows you override anything contained in this file for that particular template. As such, we won't detail the schema twice, but there are examples contained here showing how to override default configuration in your templates.
Instant Datasets are unique in that they are not allowed at the root of the default config and can only be added under environment_templates. Since Instant Datasets allow you to use instances of RDS databases (often snapshots of production, but they could be snapshots of anything) having this be the default could result in unwanted behavior for you permanent environments.
Release requires you to be explicit on which template/s you would like to (by default) use Instant Datasets. Once you have created an environment you may add Instant Datasets to your environments through the Environment Configuration file if you don't want all environments of a particular type to use datasets.
Infrastructures
Infrastructure runners are specialized jobs that execute infrastructure as code plans during deployment.
The example below shows two infrastructure runners:
Ingresses
Ingress settings that can control the behavior and functionality of the NGINX ingress controller to access HTTP services in your cluster
Example proxy buffer settings for large web requests
Example settings for stickiness settings using a cookie
Example settings for applying a WAF ruleset to the ALB in (AWS-only)
Ingress settings schema
affinity
Type of the affinity, set this to cookie to enable session affinity. See https://kubernetes.github.io/ingress-nginx/examples/affinity/cookie/
affinity_mode
The affinity mode defines how sticky a session is. Use balanced to redistribute some sessions when scaling pods or persistent for maximum stickiness.
backend_protocol
Which backend protocol to use (defaults to HTTP, supports HTTP, HTTPS, GRPC, GRPCS, AJP and FCGI)
proxy_body_size
Sets the maximum allowed size of the client request body.
proxy_buffer_size
Sets the size of the buffer used for reading the first part of the response received from the proxied server. This part usually contains a small response header.
proxy_buffering
Enables or disables buffering of responses from the proxied server.
proxy_buffers_number
Sets the number of the buffers used for reading the first part of the response received from the proxied server.
proxy_connect_timeout
Sets the timeout in seconds for establishing a connection with a proxied server or a gRPC server. Most CDN and loadbalancer timeout values are set to 60 seconds, so you should use a value like 50 seconds to ensure you do not leave stranded connections. It should be noted that this timeout cannot exceed 75 seconds.
proxy_max_temp_file_size
When buffering of responses from the proxied server is enabled, and the whole response does not fit into the buffers set by the proxy_buffer_size and proxy_buffers directives, a part of the response can be saved to a temporary file. This directive sets the maximum size of the temporary file.
proxy_read_timeout
Sets the timeout in seconds for reading a response from the proxied server. The timeout is set only between two successive read operations, not for the transmission of the whole response.
proxy_send_timeout
Sets the timeout in seconds for reading a response from the proxied server. The timeout is set only between two successive read operations, not for the transmission of the whole response.
session_cookie_change_on_failure
When set to false nginx ingress will send request to upstream pointed by sticky cookie even if previous attempt failed. When set to true and previous attempt failed, sticky cookie will be changed to point to another upstream.
session_cookie_max_age
Time in seconds until the cookie expires, corresponds to the Max-Age cookie directive.
session_cookie_name
Name of the cookie that will be created (defaults to INGRESSCOOKIE).
session_cookie_path
Path that will be set on the cookie (required because Release Ingress paths use regular expressions).
wafv2_acl_arn
The ARN for an existing WAF ACL to add to the load balancer. AWS-only, and must be created separately.
ip_allow_list
An array of source client CIDRs to allow access (e.g. 10.0.0.0/24, 172.10.0.1) at the ingress.
See https://kubernetes.github.io/ingress-nginx/user-guide/nginx-configuration/annotations/#whitelist-source-range
ip_deny_list
An array of source client CIDRs to deny access (e.g. 10.0.0.0/24, 172.10.0.1) at the ingress.
See https://kubernetes.github.io/ingress-nginx/user-guide/nginx-configuration/annotations/#denylist-source-range
Jobs
Jobs allow you to run arbitrary scripts during a deployment. This allows you to do anything before or after a service is deployed that is needed to setup your environment. A common example is database migrations before your backend comes up, but after you have deployed your database. Another good example might be running asset compilation. These tasks and any others can be accomplished using jobs.
Each job entry has a mutually exclusive requirement where either image or from_services must be present.
Jobs Example
completions
Integer amount greater than zero to indicate how many successful runs should be considered finished. Usually you would set this value equal or greater than parallelism but it might be possible to set it less if you do not care about wasted pods being scheduled. Depending on parallelism and concurrency_policy, the combination of settings may cause jobs to be run multiple times in excess of this value. See the Kubernetes documentation
parallelism
Integer amount of number of pods that can run in parallel. Set to 0 to disable the job. See the Kubernetes documentation. This controls how many pods are potentially running at the same time during a scheduled run.
service_account_name
See service accounts
node_selector
Node Selectors allow you to assign workloads to particular nodes based on common labels such as kubernetes.io/os=windows and kubernetes.io/arch=arm64. Click here for more information.
workspaces
Click here for more info on how to attach specific workspaces to this service.
Service Command
Pods running in Kubernetes typically have a default command that is run upon container start. The value specified in the command will override the supplied Docker ENTRYPOINT. Please note there is a lot of confusion for Kubernetes commands and Docker commands since they use similar and overlapping meanings for each. The command is specified as an EXECV array, not a string.
You can specify the override command that a shell will start with.
Notifications
Notifications allows you to configure how and where we notify you of events for an environment. The events for 'build', 'deploy', and 'pull_request' subscribe to all variants for those events. If you want only a message when a build fails use "build.errored".
Here are the supported values:
build - all build events
build.started - only build started events
build.completed - only build completed events
build.errored - only build errored events
deploy - all deploy events
deploy.started - only deploy started events
deploy.completed - only deploy completed events
deploy.errored - only deploy errored events
pull_request - all pull_request events
pull_request.created - only pull_request created events
NOTE: Slack integration must be configured to send any notifications to Slack.
Notifications Example
Parameters
Parameters allow you to change inputs to your environments or jobs when deploying. These parameters give you deploy time customization of your deployments giving you flexibility and simplifying you configurations.
Examples of how to using parameters are below
Example: Various parameters including string and text types. Also, it shows how to use the advanced flag.
type
The allowed types are: string, integer, list, text, boolean, git-repo, git-ref, and file
Resources
Resources are service level defaults. They represent the resources allocated for each service. Storage is different in that not every container needs storage, so while you can specify defaults, not every container will use storage.
Requests define resource guarantees. Containers are guaranteed the request amount of resource. If not enough resources are available the container will not start.
Limits, on the other hand, make sure a container never goes above a certain amount of resource. The container is never allowed to exceed the limit.
memory: Limits and requests for memory are measured in bytes. You can express memory as a plain integer or as a fixed-point integer using one of these suffixes: E, P, T, G, M, K. You can also use the power-of-two equivalents: Ei, Pi, Ti, Gi, Mi, Ki. For example, the following represent roughly the same value: 128974848, 129e6, 129M, 123Mi
nvidia_com_gpu: Limits for Nvidia GPU units. (Do not specify requests:). GPU limits can only be integer values and cannot be shared concurrently with other containers. You must also specify a node_selector to schedule a job or service on the correct worker node(s).
cpu: Limits and requests for cpu are represented in millicpu. This is represented by '{integer}m', e.x: 100m (guarantees that service will receive 1/10 of 1000m, or 1/10 of 1 cpu). You can also represent the cpu resources as fractions of integers, e.x. 0.1, is equivalent to 100m. Precision finer than '1m' is not allowed.
replicas: This number is the number of containers that will run during normal operation. This field is an integer, e.x: 5, that would make 5 of each service.
storage: Consists of two values size and type. Size accepts the same values as memory and type is the type of storage, whether persistent/nfs, empty_dir, shmem, or host_path.
Service Accounts
Service accounts allow you to control the cloud permissions granted to your workloads (services, jobs, infrastructure runners, etc.)
To apply a service account to a workload, set its service_account_name field to its name.
Example: Assuming a custom IAM role from a job
Example: Cloud account ID expansion
Services
Services contain descriptions of each of your containers. They include many fields from your docker-compose and fields auto-generated by Release upon application creation. For each service you can define:
Static javascript builds
Open and map any number of ports
Creates mounts and volumes
Use ConfigMaps to modify config at run-time for off-the-shelf containers
Override default resources
Pin particular services to particular images
Create liveness and readiness probes and set other k8s config params (e.x. max_surge)
Create stateful services
External DNS entries for cross namespace services
You can define how to build and deploy a static service or a Docker container not defined in your docker-compose file, for example, here we define a JavaScript static build:
service_account_name
See service accounts
autoscale
Read more details at the Horizontal Pod Autoscaler docs. The following example shows how you can scale based on both CPU and Memory targets:
You can also specify a custom metric that is reported by your own resource or helm chart. This is the pods metric specification which can scale pods based on some custom metric that is aggregated across the pods in question. For example if the packets_per_second metric is available in your cluster and it exceeds an average value across several pods then a scale up will be triggered. You can combine them with other metrics including CPU and memory as above.
configmap
You can use configmaps to mount a file from your repository into a path specified inside the container. You can only reference files this way, not directories.
loadbalancer
The following example shows explicitly set default values for a public node_port loadbalancer with a hostname:
The following example shows how to create a http internal loadbalancer with a hostname with TLS offload:
One of hostname (single) or hostnames (array) is the only required field to set the alias DNS for the loadbalancer as a list or single string
type supports one of layer4 (default, TCP or UDP), layer7 (optional generic passthrough), http (legacy HTTP/1.1 or HTTP/2), http2, or grpc (HTTP/2, especially the alias GRPC). Depending on your cloud provider and custom annotations, these settings may be identical.
visibility supports one of public-direct (default), or private for internet-facing or internal VPC addressing, respectively. (To be compatible with visibility rules, you can use public and unmanaged as synonyms for public-direct. The latter will not create a DNS entry.)
tls_enabled is one of true (enabled TLS negotiation and certificate offload) or false (default, for no TLS). You can also provide custom annotations to expand the coverage of the TLS configuration options.
backend_protocol is one of tcp (default, no TLS) or tls (to enable end-to-end encryption)
annotations is a hash of custom annotations to apply to the service. Depending on your cloud provider(s), these may not be portable between controllers. Mixing two cloud providers' annotations is not tested.
node_selector
Node Selectors allow you to assign workloads to particular nodes based on common labels such as kubernetes.io/os=windows and kubernetes.io/arch=arm64. Click here for more information.
nvidia_com_gpu
limits: must be an integer value. Do not specify requests:. GPU processors cannot be overprovisioned or shared with other containers.
sidecars
For more info on sidecars click here
workspaces
Click here for more info on how to attach specific workspaces to this service.
Stateful Sets and Deployments
stateful provides a StatefulSet which creates guarantees about the naming, ordering and uniqueness of a service.
Stable, unique network identifiers.
Stable, persistent storage.
Ordered, graceful deployment and scaling.
Ordered, automated rolling updates.
If an application doesn’t require any stable identifiers or ordered deployment, deletion, or scaling, you should either set stateful to false or remove it.
Build
Instructions for Release to build an image. This needs to be combined with has_repo: true.
You can specify parameters that affect how your docker image is built.
Build Image Scan
Release allows for scanning your images for vulnerabilities. If any are found, the build is marked as an error. You are able to designate what level of severity will cause an error and also whitelist specific vulnerabilities to ignore.
Example image scan that will fail the build if any CVE's with a severity level of high are found. The scan also skips over CVE-123 because it is known that Release created that fake CVE in this documentation.
Service Command
Pods running in Kubernetes typically have a default command that is run upon container start. The value specified in the command will override the supplied Docker ENTRYPOINT. Please note there is a lot of confusion for Kubernetes commands and Docker commands since they use similar and overlapping meanings for each. The command is specified as an EXECV array, not a string.
You can specify the override command that a shell will start with.
Service Resources
Resources can be overwritten on a service by service basis. The resources key is removed and each directive cpu, memory, storage, and replicas can be defined individually. If they are not specified the defaults will be used.
cpu, memory, nvidia_com_gpu, and storage define resource guarantees. The service definition for cpu, memory, nvidia_com_gpu, and storage overrides the values in resource_defaults. In the case of nvidia_com_gpu, Kubernetes recommends setting limits: but not requests:, unless they are the same. You can use the service definition to more finely tune the amount cpu, memory, nvidia_com_gpu, and storage for each service.
replicas allows you to specify different amount of pods that will be deployed for your particular service.
Init Containers
Init containers allow you define additional containers that share volume mounts from the primary service. These can be used to perform setup tasks that are required for the main service to run. Init containers should run to completion with an exit code of zero. Non zero exit codes will result in a CrashLoopBackoff.
Example init container which inherits image from the main service
You can also define init containers using off the shelf images like busybox. This can be useful for performing additional operations which don't require the main service image or require binaries not included in the primary service.
Example init container using busybox to wait for another service to startup
volumes
See Volumes
Service Command
Pods running in Kubernetes typically have a default command that is run upon container start. The value specified in the command will override the supplied Docker ENTRYPOINT. Please note there is a lot of confusion for Kubernetes commands and Docker commands since they use similar and overlapping meanings for each. The command is specified as an EXECV array, not a string.
You can specify the override command that a shell will start with.
Startup, Readiness, and Liveness Probes
startup_probe, liveness_probe, and readiness_probe are used to check the health of your service. When your code is deployed via a rolling deployment, the readiness_probe will determine if the service is ready to serve traffic before adding it to the load balancer. Release will convert the docker-compose healthcheck to a liveness_probe and readiness_probe. Both liveness_probe and readiness_probe allow for more advanced configuration beyond the docker-compose healthcheck definition. A startup_probe may be used if an application takes a long time during startup.
In this example we show the various types of probes that you can define for services along with overrides for resources and timeouts, while also defining static builds.
Service Node Selectors
Node Selector allows pods to chose specific nodes to run on. The most common use case is for selecting nodes with different OS (like Windows) or different architecture (like ARM64, GPUs), but could also select specific cloud provider settings such as AWS AZ (like us-east-1c).
Example Windows 2019 server selector for a specific zone
Example Nvidia server selector for GPU processors
Notice
Please note the top-level node_selector will be treated as a global default that can be overridden inside each service or job.
Automatic tolerations
Release will automatically apply tolerations for the following well-known selectors:
kubernetes.io/os=windowskubernetes.io/arch=arm64nvidia.com/gpu="true"
This allows linux/amd64 workloads to mix with Windows, ARM64, and GPU nodegroup workloads.
Helm examples
Release does not manage any Helm charts to use node_selectors, so you will need to add your own YAML and tolerations as follows for Windows as an example:
Example Helm chart addition to schedule Windows nodes.
Example Helm chart addition to schedule GPU nodes.
Example Helm chart addition to schedule ARM64 nodes.
Further Reading
Kubernetes Taints Documentation Kubernetes GPU Documentation
key
The kubernetes label key to select on.
value
The value of the kubernetes label to match on.
Ports
Ports can be one of two types container_port or node_port.
container_port is used to define a port that another service will consume. Internal services like your data stores, caches and background workers should not be exposed to the internet and available only internally to other services.
node_port is used to define a service that you want to expose to the Internet. target_port is the port on the pod that the request gets sent to from a load balancer or internally. Your application needs to be listening for network requests on this port for the service to work. port exposes the service on the specified port internally within the cluster.
You can set an optional loadbalancer flag to create a separate load balancer that can be used access the service over the Internet. loadbalancer is useful for exposing a TCP based service to the Internet that doesn't support HTTP/HTTPS traffic.
node_port will also define an ingress rule to allow HTTP/HTTPS traffic to be routed to a service. See the documentation on hostnames and rules to understand how to define ingress rules and mount your service at a custom path, etc.
Container and Node Ports together
Secrets
You can use secrets to mount a file with the contents of a secret into a path specified inside the container.
Each entry in the secrets stanza requires a mount path, which needs to be an unused directory. For each mount path, a list of items is used to populate the files with the secret values. The value must be sourced from Release's Secrets Manager. The following YAML results in two secrets being mounted, one at /etc/foo/my-secrets/my-secret.conf and another at /run/secrets/db-password.
As noted above, the mount path has to be an unused directory. If you need the file to be mounted into a directory that is already populated, we suggest moving the files in the startup command for the service. An example is shown below moving the files inline, however this same result could be achieved using a script.
Secret Items
Attributes of the items in the secrets attribute
Storage
Usage of storage is deprecated See volumes and use size. If storage is present as well as size under volumes, thevolumes attributes will take precedence
Use storage in combination with volumes to create persistent storage for your service.
Example storage definition with volumes to retain Postgresql data on container restarts
size
Measured in bytes. Defaults to 1Gi. You can express size as a plain integer or as a fixed-point integer using one of these suffixes: E, P, T, G, M, K. You can also use the power-of-two equivalents: Ei, Pi, Ti, Gi, Mi, Ki. For example, the following represent roughly the same value: 128974848, 129e6, 129M, 123Mi
Volumes
Volumes provide a means to storing information on disk. This ensures no data is lost if the Pod restarts or as a means to sharing files between Pods.
Further reading can be found on the Kubernetes documentation for Volumes
Example
persistentvolume to retain Postgresql data on container restarts
Example
empty_dirvolume
Example
shmemvolume
Example
host_pathvolume
Example
s3volume
claim
See Shared Volumes for more information.
name
type: persistent -
nameis only reference for your team. Release auto generates the name of the PersistentVolumeClaimtype: empty_dir - The name is added to Kubernetes
type: shmem - The name is added to Kubernetes
type: host_path - The name is added to Kubernetes
path
This field is deprecated, please do not use. It will be removed in future versions.
server
This field is deprecated, please do not use. It will be removed in future versions.
size
See the Kubernetes documentation for Memory Resource Units
type
The allowed types
persistent - Creates a PersistentVolumeClaim and a PersistentVolume. See the Kubernetes documentation for Persistent Volumes
empty_dir - An emptyDir volume is initially empty. See the Kubernetes documentation for emptyDir
shmem - An emptyDir volume backed by memory. The memory is deducted from the container memory. You must set requests/limits appropriately. See the empty_dir documentation above.
host_path - A hostPath volume mounts a file or directory from the host node's filesystem. See the Kubernetes documentation for hostPath
s3 - An S3 volume present in the
s3_volumesconfiguration. See s3_volumes for information on how to create them.
Jobs and Services Workspace Mounts
Jobs and Services can specify workspaces they would like mounted explicitly instead of using auto_mount. Auto mount will mount the workspace in every container and job. If you would rather control which workspaces are mounted in which jobs and/or services use the workspaces key word in your service and/or job definition is a way to do that. You are able to reference a workspace and and create a unique mount path for each job and/or service.
If you would like to read about workspaces in general click here.
Examples:
Attach workspace: ws3 to service: svc1 and ws2 to job: echo. Workspace: ws1 is attached to all jobs and services
Shared Volumes
Shared Volumes are optional. Each Shared Volume creates a single PersistentVolumeClaim and PersistentVolume. To connect a service to the shared volume, define a volume which sets its claim attribute to the name attribute of the shared volume.
Example of shared volumes using persistent
path
This field is deprecated, please do not use. It will be removed in future versions.
server
This field is deprecated, please do not use. It will be removed in future versions.
size
See the Kubernetes documentation for Memory Resource Units
type
The only currently supported value is persistent. Shared volumes are implemented using NFS in the cloud provider.
Sidecars
It’s a generally accepted principle that a container should address a single concern only. Sidecar containers allow you to define additional workloads for a given service. Sidecar containers share resources with the main service container so they are great for shared filesystems, networking ports, routing process spaces, etc..
Example sidecar definition with a reusable logstash container and nginx for serving static assets
Service Command
Pods running in Kubernetes typically have a default command that is run upon container start. The value specified in the command will override the supplied Docker ENTRYPOINT. Please note there is a lot of confusion for Kubernetes commands and Docker commands since they use similar and overlapping meanings for each. The command is specified as an EXECV array, not a string.
You can specify the override command that a shell will start with.
Workflows
By default there are three workflows : setup, patch, and teardown. setup is what creates your environment from scratch. patch is the workflow for deploying your code to an already setup environment. teardown is the workflow for deleting your environment and removing any cloud native resources that have been created during setup. These are auto-generated, but you can add your own jobs and change the order, add tasks, etc.
You may define your own workflows that you run on a one-off basis. These workflows allow you to run jobs inside the namespace of the deployed environment. Whenever, you click the Deploy button you will be given the option to run your one-off workflows.
wait_for_all_tasks_to_complete is set to true by default. This means that when the deployment finishes it will wait for all the tasks to finish. If it's set to false, the stage will not wait for everything to finish and the next stage will run immediately.
Example: Example showing setup, patch, teardown and a custom workflow that ingests docs from gitbook
Workflow Parallelization
Parallelization of your workflows can result in significant decrease in the time it takes to deploy your environment. But, it may not be as simple as just telling Release to run all of your services and/or jobs in parallel. In some cases you may need certain services to wait before a migration job has run, for example. You can design your application around this issue, but in some cases it will be unavoidable and parallelize gives you both options.
wait_for_finish allows you to customize the behavior of each step. By default everything in a step is run in parallel, but the workflow will not transition to the next step until every task in the step has finished. By setting it to false, the workflow will start it and the progress to the next step.
The most obvious use-case for it being false, is a long running frontend static build. You want to start it at the beginning of the workflow and make wait_for_finish: false, this way it will not hold up things that don't depend on it and by default the workflow will wait at the end for all jobs to finish, before transitioning to the next workflow.
metadata allows you to customize tasks by passing directives or data to them. It allows you to have multiple tasks in your task list, but send different kinds of data or params to specific tasks. See the schema for metadata in Release Tasks
Example: Same services and jobs, but the frontend will be started first, but everything else will be allowed to run without waiting for it to finish. Only after everything else is done, will the workflow wait for
services.frontendto finish.
Release Tasks (beta)
Release has created a few tasks that you can reference and use in your workflows. These tasks are built into release and you can parameterize them and utilize them in your workflows.
pod_exec: This task allows you to run any arbitrary commands on the pods of your choosing. This is very useful for sending messages to the service/s you have running on each pod. A good use case is you need to send a signal to all of your queue workers. If you used aJob(k8s jobs) you would need to find all the pods, etc to run the command on. This way Release does that for you and executes your command on each Pod.pod_checker: This task will check the states of the pods and not finish until at least one of those states is found on every pod. This is very useful if you would like to do something to, or run something on, or against those pods, but only after they have transitioned to a specific state or one of a list of states.remove_environment: This task is required in theteardownworkflow. When it runs, Release will remove your environment from the UI and remove the namespace along with all corresponding objects from Kubernetes.
Pod Exec Metadata Schema
When you are using metadata to parameterize your pod_exec job, only a subset of the metadata directives are available.
Pod Exec Example
Pod Exec Example: We are going to
ruby ./bin/backend_pod_setup.rbon each pod for the backend service.
Full Metadata Schema
More Examples
Pod Checker and Pod Exec Combined Example: We are checking the backend pods to make sure they are in the
runningstate before runningruby ./bin/backend_pod_setup.rbon each pod.
Workspaces
Workspaces are collections of data sources and/or repositories that are made available for your containers. Workspaces allow you access to the data you need for your workflows located locally or in the cloud (S3, Cloud Storage, etc). Release can mount S3 buckets, Cloud Storage Buckets, and Git Repositories.
You can make workspaces more dynamic by combining Parameters and Workspaces to change mounts or paths at deploy time.
Examples:
Mount github repo at /data in the workspace and an S3 bucket in second workspace
Use parameters to change source URL upon environment creation
Workspace Mounts
A list of mount paths and source urls for datasources you would like mounted in your containers and/or jobs.
Examples:
Mount github repo at /data in the workspace and an S3 bucket in second workspace
Last updated