The plugin requires Serverless 1.12 or later!
This plugin enables use of AWS aliases on Lambda functions. The term alias must not be mistaken as the stage. Aliases can be deployed to stages, e.g. if you work on different VCS branches in the same service, you can deploy your branch to a new alias. The alias deployment can contain a different set of functions (newly added ones or removed ones) and does not impact any other deployed alias. Aliases also can be used to provide a 'real' version promotion.
As soon as the service is deployed with the plugin activated, it will create a default alias that is named equally to the stage. This is the master alias for the stage.
Each alias creates a CloudFormation stack that is dependent on the stage stack. This approach has multiple advantages including easy removal of any alias deployment, protecting the aliased function versions, and many more.
Add the plugin to your package.json's devDependencies and to the plugins array
in your serverless.yml file. After installation the plugin will automatically
hook into the deployment process.
Additionally the new alias command is added to Serverless which offers some
functionality for aliases.
The default alias (for the stage) is deployed just by doing a standard stage
deployment with serverless deploy. From now on you can reference the aliased
versions on Lambda invokes with the stage qualifier. The aliased version is
read only in the AWS console, so it is guaranteed that the environment and
function parameters (memory, etc.) cannot be changed for a deployed version
by accident, as it can be done with the $LATEST qualifier.
This adds an additional level of stability to your deployment process.
To deploy an alias to a stage, just add the --alias option to serverless deploy
with the alias name as option value.
Example:
serverless deploy --alias myAlias
See the alias remove command below.
To remove a complete service, all deployed user aliases have to be removed first,
using the alias remove command.
To finally remove the whole service (same outcome as serverless remove), you have
to remove the master (stage) alias with serverless alias remove --alias=MY_STAGE_NAME.
This will trigger a removal of the master alias CF stack followed by a removal of the service stack. After the stacks have been removed, there should be no remains of the service.
The plugin will print reasonable error messages if you miss something so that you're guided through the removal.
In Serverless stages are, as above mentioned, parallel stacks with parallel resources. Mapping the API Gateway resources to this semantics, each stage has its own API deployment.
Aliases fit into this very well and exactly as with functions an alias is a kind of "tag" within the API deployment of one stage. Curiously AWS named this "stage" in API Gateway, so it is not to be confused with Serverless stages.
Thus an alias deployment will create an API Gateway stage with the alias name as name.
The created API Gateway stage has the stage variables SERVERLESS_STAGE and SERVERLESS_ALIAS set to the corresponding values.
If you want to test your APIG endpoints in the AWS ApiGateway console, you have to set the SERVERLESS_ALIAS stage variable to the alias that will be used for the Lambda invocation. This will call the aliased function version.
Deployed stages have the alias stage variable set fixed, so a deployed alias stage is hard-wired to the aliased Lambda versions.
The alias plugin supports configuring the deployed API Gateway stages, exactly as you can do it within the AWS APIG console, e.g. you can configure logging (with or without data/request tracing), setup caching or throttling on your endpoints.
The configuration can be done on a service wide level, function level or method level
by adding an aliasStage object either to provider, any function or a http event
within a function in your serverless.yml. The configuration is applied hierarchically,
where the inner configurations overwrite the outer ones.
HTTP Event -> FUNCTION -> SERVICE
All settings are optional, and if not specified will be set to the AWS stage defaults.
aliasStage:
cacheDataEncrypted: (Boolean)
cacheTtlInSeconds: (Integer)
cachingEnabled: (Boolean)
dataTraceEnabled: (Boolean) - Log full request/response bodies
loggingLevel: ("OFF", "INFO" or "ERROR")
metricsEnabled: (Boolean) - Enable detailed CW metrics
throttlingBurstLimit: (Integer)
throttlingRateLimit: (Number)
There are two further options that can only be specified on a service level and that affect the whole stage:
aliasStage:
cacheClusterEnabled: (Boolean)
cacheClusterSize: (Integer)
For more information see the AWS::APIGateway::Stage or MethodSettings documentation on the AWS website.
Sample serverless.yml (partial):
service: sls-test-project
provider:
...
# Enable detailed error logging on all endpoints
aliasStage:
loggingLevel: "ERROR"
dataTraceEnabled: true
...
functions:
myFunc1:
...
# myFunc1 should generally not log anything
aliasStage:
loggingLevel: "OFF"
dataTraceEnabled: false
events:
- http:
method: GET
path: /func1
- http:
method: POST
path: /func1/create
- http:
method: PATCH
path: /func1/update
# The update endpoint needs special settings
aliasStage:
loggingLevel: "INFO"
dataTraceEnabled: true
throttlingBurstLimit: 200
throttlingRateLimit: 100
myFunc2:
...
# Will inherit the global settings if nothing is set on function level
You can reference the currently deployed alias with ${self:provider.alias} in
your service YAML file. It should only be used for information, but not to
set any resource names. Making anything hard-wired to the alias name might
make the project unusable when deployed to different aliases because the resources
are maintained in the master CF stack - the plugin takes care that resources
are available.
A valid use is to forward the alias name as environment variable to the lambdas and use it there for tagging of log messages. Then you see immediately which aliased lambda is the origin.
Any other use with the further exception of lambda event subscriptions (see below) is strongly discouraged.
Resources are deployed per alias. So you can create new resources without destroying the main alias for the stage. If you remove an alias the referenced resources will be removed too.
However, logical resource ids are unique per stage. If you deploy a resource into one alias, it cannot be deployed with the same logical resource id and a different configuration into a different alias. Nevertheless, you can have the same resource defined within multiple aliases with the same configuration.
This behavior exactly resembles the workflow of multiple developers working on different VCS branches.
The master alias for the stage has a slightly different behavior. If you deploy here, you are allowed to change the configuration of the resource (e.g. the capacities of a DynamoDB table). This deployment will reconfigure the resource and on the next alias deployment of other developers, they will get an error that they have to update their configuration too - most likely, they updated it already, because normally you rebase or merge your upstream and get the changes automatically.
Event subscriptions that are defined for a lambda function will be deployed per alias, i.e. the event will trigger the correct deployed aliased function.
Subscriptions to SNS topics can be implicitly defined by adding an sns event to
any existing lambda function definition. Serverless will create the topic for you
and add a subscription to the deployed function.
With the alias plugin the subscription will be per alias. Additionally the created topic is renamed and the alias name is added (e.g. myTopic-myAlias). This is done because SNS topics are independent per stage. Imagine you want to introduce a new topic or change the data/payload format of an existing one. Just attaching different aliases to one central topic would eventually break the system, as functions from different stages will receive the new data format. The topic-per-alias approach effectively solves the problem.
If you want to refer to the topic programmatically, you just can add -${process.env.SERVERLESS_ALIAS}
to the base topic name.
Event subscriptions can reference resources that are available throughout all aliases if they reference the same resource id. That means that an event will trigger all aliases that are deployed with the subscription defined.
Example:
functions:
testfct1:
description: 'My test function'
handler: handlers/testfct1/handler.handle
events:
- stream:
type: kinesis
arn: "arn:aws:kinesis:${self:provider.region}:XXXXXX:stream/my-kinesis"
- http:
method: GET
path: /func1
resources:
Resources:
myKinesis:
Type: AWS::Kinesis::Stream
Properties:
Name: my-kinesis
ShardCount: 1
When a function is deployed to an alias it will now also listen to the my-kinesis stream events. This is useful, if you want to test new implementations with an existing resource.
Currently this feature is not available. The Serverless framework does not support variable substitution in property names (see #49). As soon as this has been implemented there, this note will be removed.
There might be cases where you want to test with your private resources first, before you deploy changes to the master alias. Or you just want to create separate resources and event subscriptions per alias.
The solution here is to make the resource id dependent on the alias name, so that the alias effectively owns the resource and the event subscription is bound to it.
Example:
functions:
testfct1:
description: 'My test function'
handler: handlers/testfct1/handler.handle
events:
- stream:
type: kinesis
arn: "arn:aws:kinesis:${self:provider.region}:XXXXXX:stream/my-kinesis-${self.provider.alias}"
- http:
method: GET
path: /func1
resources:
Resources:
myKinesis${self:provider.alias}:
Type: AWS::Kinesis::Stream
Properties:
Name: my-kinesis-${self.provider.alias}
ShardCount: 1
The examples above use named streams. I know that this is not perfect as changes that require replacement are not possible. The reason for the named resources is, that Serverless currently only supports event arns that are strings. The change is already in the pipeline there. Afterwards you just can reference the event arns with CF functions like "Fn::GetAtt" or "Ref". I will update the examples as soon as it is fixed there and publicly available.
The plugin integrates with the Serverless info command. It will extend the information that is printed with the list of deployed aliases.
In verbose mode (serverless info -v) it will additionally print the names
of the lambda functions deployed to each stage with the version numbers the
alias points to.
Given an alias with --alias=XXXX info will show information for the alias.
The plugin integrates with the Serverless logs command (all standard options will
work). Additionally, given an alias with --alias=XXXX, logs will show the logs
for the selected alias. Without the alias option it will show the master alias
(aka. stage alias).
Removes an alias and all its uniquely referenced functions and function versions.
The alias name has to be specified with the --alias option.
Functions and resources owned by removed aliases will be physically removed after the alias stack has been removed.
The alias plugin is compatible with all standard Serverless commands and switches.
For example, you can use --noDeploy and the plugin will behave accordingly.
Care has to be taken when using other plugins that modify the CF output too. I will add configuration instructions in this section for these plugin combinations.
The warmup plugin will keep your Lambdas warm and reduce the cold start time effectively. When using the plugin, it must be listed before the alias plugin in the plugin list of serverless.yml. The warmup lambda created by the plugin will be aliased too, so that the warmup plugin can be configured differently per deployed alias.
In case you wanna test how it behaves, I provided a predefined test service in
the sample folder, that creates two functions and a DynamoDB table.
Feel free to check everything - and of course report bugs immediately ;-)
as I could not test each and every combination of resources, functions, etc.
A common use case is that multiple developers work on different branches on the same service, e.g. they add new functions individually. Every developer can just deploy his version of the service to different aliases and use them. Neither the main (stage) alias is affected by them nor the other developers.
If work is ceased on a branch and it is deleted, the alias can just be removed and on the next deployment of any other alias, the obsoleted functions will disappear automatically.
A preconfigured sample project can be found here. It lets you start testing right away. See the project's README for instructions. The sample project will evolve over time - when new features or changes are integrated into the plugin.
If you are not happy with the plugin or just do not like me, you can easily get rid of the plugin without doing any harm to the deployed stuff. The plugin is non-intrusive and does only add some output variables to the main stack:
- Remove all alias stacks via the CloudFormation console or with 'alias remove'
- Remove the plugin from your serverless.yml and your package.json
- Deploy the service again (serverless deploy)
You're all set.
Aliases can have different VPC settings (serverless.yml:provider.vpc). So you can use an alias deployment also for testing lambda functions within other internal networks. This is possible because each deployed AWS lambda version contains its entire configuration (VPC settings, environment, etc.)
currently the exposed hooks are not available after the change to the new SLS lifecycle model
The plugin adds the following lifecycle events that can be hooked by other plugins:
-
alias:deploy:uploadArtifacts
Upload alias dependent CF definitions to S3.
-
alias:deploy:updateAliasStack
Update the alias CF stack.
-
alias:deploy:done
The Alias plugin is successfully finished. Hook this instead of 'after:deploy:deploy' to make sure that your plugin gets triggered right after the alias plugin is done.
-
alias:remove:removeStack
The alias stack is removed from CF.
If you hook after the alias:deploy:loadTemplates hook, you have access to the currently deployed CloudFormation templates which are stored within the global Serverless object: serverless.service.provider.deployedCloudFormationTemplate and serverless.service.provider.deployedAliasTemplates[].
- The master alias for a stage could be protected by a separate stack policy that only allows admin users to deploy or change it. The stage stack does not have to be protected individually because the stack cross references prohibit changes naturally. It might be possible to introduce some kind of per alias policy.
-
1.3.0
-
1.2.1
-
1.2.0
-
1.1.0
-
1.0.0 Support "serverless logs" with aliases. First non-alpha!