awslim
is a CLI for AWS services by Go. This CLI is generated from the AWS SDK Go v2 service client.
(The old name is aws-sdk-client-go
.)
While the AWS CLI is very useful, it can be resource intensive to boot up. awslim
offers a simpler and faster alternative for limited use cases. It acts as a simple wrapper around AWS SDK Go v2, providing essential functionality without the full feature set of the AWS CLI.
- Call any method(API) of the AWS service client.
- Use JSON or Jsonnet for input.
- Output the result in JSON format.
- Bind a file to the input/output struct of the method.
- Query the output by JMESPath.
- Use the AWS CLI configuration file. (i.e.,
~/.aws/config
)
- Not 100% compatible with the AWS CLI.
- No support for AWS CLI plugins. (i.e.,
session-manager-plugin
) - Some service names and different from the AWS CLI. (i.e.,
logs
->cloudwatchlogs
,ce
->costexplorer
)
Note: The release binaries are large (about 500MB after being extracted) and have a slight startup delay (about 100ms), due to the inclusion of code to access all AWS services. However, it is still faster than the AWS CLI.
For optimized performance, build your own binary tailored to the specific services you need. See the Build section for details.
Download the binary from the release page.
$ brew install fujiwara/tap/awslim
You can build the client yourself, including only the needed services and methods. This produces a smaller, faster binary.
The client is built by using a AWSLIM_GEN
environment variable or a gen.yaml
configuration file.
Set the AWSLIM_GEN
environment variable to specify the services you want to include, separated by commas.
For example, to build the client for ECS, Firehose, and S3:
$ export AWSLIM_GEN="ecs,firehose,s3"
This will generate all methods for the specified services. If you want to generate only specific methods, use the gen.yaml configuration file.
# gen.yaml
services:
ecs:
- DescribeClusters
- DescribeTasks
firehose:
- DescribeDeliveryStream
- ListDeliveryStreams
s3:
# all methods of the service
Keys under services
are AWS service names (github.com/aws/aws-sdk-go-v2/service/*
), and values are method names of the service client (for example, s3
is s3.Client). If you don't specify the method names, all methods of the service client are generated.
Set the AWSLIM_OS
and AWSLIM_ARCH
environment variables to specify the OS and architecture for the build.
If these variables are not set, awslim will build based on the architecture of the build environment.
These variables can also be specified when building with Docker.
The values that can be set for AWSLIM_OS and AWSLIM_ARCH are the same as for GOOS and GOARCH.
To build the client, run the following commands (or simply run make
):
$ go generate ./cmd/awslim-gen .
$ go build -o your-client ./cmd/awslim/main.go
go generate ./cmd/awslim-gen .
generates the generator bygen.yaml
.go build -o your-client ./cmd/awslim/main.go
builds your client.
If you change the configuration, run make clean
before make
to purge the generated files.
Use the ghcr.io/fujiwara/awslim:builder
builder image to build the client inside a container.
Environment variables:
GIT_REF
: Git reference to checkout the repository. Default ismain
. You can specify a branch, tag, or commit hash.
Example using the AWSLIM_GEN
environment variable:
$ docker run -it -e AWSLIM_GEN=ecs,firehose,s3 ghcr.io/fujiwara/awslim:builder
...
Example to specify build for OS and Architecture:
$ docker run -it -e AWSLIM_GEN=ecs,firehose,s3 -e AWSLIM_OS=linux -e AWSLIM_ARCH=arm64 ghcr.io/fujiwara/awslim:builder
...
Example using the gen.yaml
configuration file:
$ docker run -it -v $(pwd)/gen.yaml:/app/gen.yaml ghcr.io/fujiwara/awslim:builder
...
Completed. Please extract /app/awslim from this container!
For example, run the following command:
docker cp $(docker ps -lq):/app/awslim .
After the build is complete, copy the binary to your host machine with the docker cp
command.
$ docker cp $(docker ps -lq):/app/awslim .
It is also possible to use the ghcr.io/fujiwara/awslim:builder
builder image in a multi-stage build.
Run ./build-in-docker.sh
in the container to build the client. The built binary will be located in the /app
directory. You can then copy it to the final image.
FROM ghcr.io/fujiwara/awslim:builder AS builder
ENV AWSLIM_GEN=ecs,firehose,s3
ENV GIT_REF=v0.3.0
RUN ./build-in-docker.sh
FROM debian:bookworm-slim
COPY --from=builder /app/awslim /usr/local/bin/awslim
Example of executing sts get-caller-identity
on a 0.25 vCPU Fargate(AMD64) using /usr/bin/time -v
for time measurement.
command | CPU time(user, sys) | Elapsed time(s) | Max memory(MB) | Size(MB) |
---|---|---|---|---|
aws | 0.67 + 0.10 = 0.77 | 3.11 | 64.2 | 225 |
awslim(all) | 0.08 + 0.03 = 0.11 | 0.43 | 101.5 | 476 |
awslim(40) | 0.02 + 0.01 = 0.03 | 0.05 | 30.2 | 95 |
awslim
(built for all AWS services): 7.0x faster thanaws
awslim
(built for 40 AWS services): 62.0x faster thanaws
aws-cli/2.15.51 Python/3.11.8
, awslim 0.1.0
Usage: awslim [<service> [<method> [<input> [<args> ...]]]] [flags]
Arguments:
[<service>] service name
[<method>] method name
[<input>] input JSON/Jsonnet struct or filename
[<args> ...] additional flags/args
Flags:
-h, --help Show context-sensitive help.
-i, --input-stream=STRING bind input filename or '-' to io.Reader field in the input struct
-o, --output-stream=STRING bind output filename or '-' to io.ReadCloser field in the output struct
--[no-]api-output output API response into stdout
-r, --raw-output output raw strings, not JSON texts
-c, --compact compact JSON output
-q, --query=STRING JMESPath query to apply to output
--ext-str=KEY=VALUE;... external variables for Jsonnet
--ext-code=KEY=VALUE;... external code for Jsonnet
--[no-]strict strict input JSON unmarshaling
-f, --follow-next="" OutputField=InputField format. follow the next token.
--camel convert keys to camelCase
-n, --dry-run dry-run mode
-v, --version show version
--debug turn on debug logging
service
: AWS service name.method
: Method name of the service client.input
: JSON input for the method.
The output is JSON format.
$ awslim
$ awslim ecs
The third argument is a JSON or Jsonnet input for the method. This can be omitted if the method requires no input ({}
is passed implicitly).
$ awslim ecs ListTasks '{Cluster:"default"}' # JSON
$ awslim ecs ListTasks '{Cluster:"default"}' # Jsonnet
If the method name is "kebab-case", it automatically converts to "PascalCase" (i.e., list-tasks
-> ListTasks
).
$ awslim ecs list-tasks '{Cluster:"default"}'
In v0.3.0, flag arguments can be specified, like the AWS CLI!
$ awslim ecs list-tasks --cluster default
Note: Currently, flag arguments do not support setting non-string fields (array, object, number, and boolean). Use JSON or Jsonnet for such fields.
$ awslim ecs list-tasks '{MaxResults:10}' --cluster default
The third argument can be a filename that contains JSON or Jsonnet input.
$ awslim ecs list-tasks input.jsonnet
Note: By default, the input JSON is unmarshaled strictly. Unknown fields for the input struct in the input JSON cause an error. If you want to unmarshal the input JSON non-strictly, use --no-strict
option.
awslim supports Jsonnet functions for the input JSON.
$ awslim ecs list-tasks '{Cluster: _(0)}' foo
$ CLUSTER=foo awslim ecs list-tasks '{Cluster: env("CLUSTER","default")}'
$ CLUSTER=foo awslim ecs list-tasks '{Cluster: must_env("CLUSTER")}'
_(n)
returns the n-th argument.env(n, d)
returns the environment variablen
. If the environment variable is not set, it returnsd
.must_env(n)
returns the environment variablen
. If the environment variable is not set, it causes an error.
When you specify the input JSON as a file, you must define the function in the Jsonnet file.
local _ = std.native('args');
local env = std.native('env');
local must_env = std.native('must_env');
{
Cluster: _(0),
}
Pass external variables to Jsonnet.
This is useful when you want to use variables in Jsonnet.
$ awslim ecs list-tasks my.jsonnet --ext-str Cluster=default
// my.jsonnet
{
Cluster: std.extVar("Cluster"),
}
Bind a file or stdin to the input struct.
$ awslim s3 put-object '{Bucket:"my-bucket",Key:"my.txt"}' --input-stream my.txt
s3#PutObjectInput has Body
field of io.Reader
. --input-stream
option binds the file to the field.
When the input struct has only one field of io.Reader
, awslim
reads the file and binds it to the field automatically. (Currently, all SDK input structs have at most one io.Reader field.)
When the input struct has a "*Length" field for the size of the content, awslim
sets the size of the content to the field automatically. For example, s3#PutObjectInput has ContentLength
field.
If --input-stream
is "-", awslim
reads from stdin. In this case, awslim
reads all contents into memory, so it is not suitable for large files. Consider using a file for large content.
Bind the io.ReadCloser
of the API output to a file or stdout.
$ awslim s3 get-object '{Bucket:"my-bucket",Key:"my.txt"}' --output-stream my.txt
s3#GetObjectOutput has Body
field of io.ReadeCloser
. --output-stream
option binds the file to the field.
When the output struct has only one field of io.ReadCloser
, awslim
copies it to the file automatically. (Currently, all SDK output structs have at most one io.ReadCloser field.)
If --output-stream
is "-", awslim
writes into stdout. The result of the API also writes to stdout by default. If you don't want to output the result, use --no-api-output
.
Output raw strings, not JSON texts.
This option is like jq -r
.
Set the output/input field name of the next token. This option is useful for paginated APIs.
For example, s3#ListObjectsV2Output has a NextContinuationToken
field, and s3#ListObjectsV2Input has a ContinuationToken
field. You can follow the next token by the following command.
{FieldInOutput}={FieldInInput}
format is used for --follow-next
option.
$ awslim s3 list-objects-v2 '{Bucket:"my-bucket"}' \
--follow-next NextContinuationToken=ContinuationToken
If the same field name is used in the output and input, you can omit the input field name.
$ awslim ecs list-tasks '{Cluster:"default"}' \
--follow-next NextToken
Convert JSON keys to camelCase.
By default, the output JSON keys are the same as the SDK struct field names (equivalent to PascalCase).
In the JSON output produced by AWS CLI, the key naming conventions are either PascalCase or camelCase, determined by the service. For example, aws ecs
uses camelCase, while aws lambda
uses PascalCase.
If you want to convert the keys to camelCase, use the --camel
option.
$ awslim ecs describe-clusters '{Clusters:["default"]}' --camel
{
"clusters": [
{
"activeServicesCount": 1,
"capacityProviders": [],
"clusterArn": "arn:aws:ecs:ap-northeast-1:123456789012:cluster/default",
"clusterName": "default",
"defaultCapacityProviderStrategy": [],
"pendingTasksCount": 0,
"registeredContainerInstancesCount": 0,
"runningTasksCount": 0,
"settings": [],
"statistics": [],
"status": "ACTIVE",
"tags": []
}
],
"failures": [],
"resultMetadata": {}
}
This conversion is performed mechanically, so objects for which any key can be specified (such as the dockerLabels element in an ECS task definition) are also subject to conversion.
It is not guaranteed that the results will match those in the AWS CLI output.
Query the output by JMESPath like the AWS CLI.
$ awslim sts get-caller-identity --query 'Account'
"012345678901"
Use the help
argument to display the URL of the method's documentation. Since awslim
is a simple wrapper for the AWS SDK Go v2 service client, its usage is the same as the SDK.
$ awslim ecs DescribeClusters help
See https://pkg.go.dev/github.com/aws/aws-sdk-go-v2/service/ecs#Client.DescribeClusters
awslim
reads the configuration file (~/.config/awslim/config.(json|jsonnet|yaml|yml)
). The configuration file can contain the following fields:
open: /usr/bin/open
aliases:
whoami: sts get-caller-identity
regions: ec2 describe-regions --query Regions[].RegionName
open
: The command to open the URL to the documentation.aliases
: The alias of the cli arguments. For example,awslim whoami
is equivalent toawslim sts get-caller-identity
.
XDG_CONFIG_HOME
environment variable is used for the configuration file path. If XDG_CONFIG_HOME
is not set, ~/.config
is used.
MIT
Fujiwara Shunichiro