See Contributing for contribution guidelines.

Setup

aws-api

git clone git@github.com:cognitect-labs/aws-api.git
cd aws-api
clj -Sdeps

Run tests

We need the dependencies in the :dev alias and the test runner in the :test or :test-integration aliases in order to run the tests.

To run the integration tests, you'll need to bind the AWS_PROFILE env var to a profile named aws-api-test, which should be configured to access a development account in order to avoid running destructive operations in a prod account.

;; unit tests
clj -M:dev:test

;; integration tests
AWS_PROFILE=aws-api-test clj -M:dev:test-integration

Add new tests

Tests live in test/src. Unit tests should be in a parallel path to the namespace you want to test, adding _test in the end of the filename. For example, test/src/cognitect/aws/credentials_test.clj contains tests for src/cognitect/aws/credentials.clj

Integration tests go in test/src/cognitect/aws/integration/. When creating integration tests, please add ^:integration in deftest. Example:

(deftest ^:integration do-something-with-a-real-aws-service
  ...)

It is also important to ensure that the AWS_PROFILE env var is the test profile aws-api-test. To do this add this at the beginning of your integration test namespace:

(ns cognitect.aws.integration.s3-test
  (:require [clojure.test :refer [use-fixtures]]
            [cognitect.aws.integration.fixtures :as fixtures]))

(use-fixtures :once fixtures/ensure-test-profile)

Build

Build and install in your local ~/.m2 (or wherever you store maven repos)

build/package

CI/CD

This repo relies on GitHub Actions for CI/CD. See the .github/workflows/ directory.

Secrets

The integration tests require an AWS access key and secret key pair to run. The Run integration tests step in the test.yml file has an env which refers to these two secrets, whose values are stored as GitHub repository secrets.

Release

The release is automated via a GitHub workflow (.github/workflows/release.yml). The workflow can be triggered on demand by an authorized user.

The release workflow makes use of the build/release script.

build/release

In summary, the release process:

• Releases are done from the main branch of the repository.
• The release artifacts are built, signed and deployed to Maven central.
• The git repository is tagged with a release tag.
• Updates CHANGES.md, README.md, latest-releases.edn
• Updates the API documentation in the gh-pages branch

pre-release checklist

• Ensure the CHANGES.md file (and, optionally, the UPGRADE.md file) contains the expected latest unreleased changes, under a heading of ## DEV near the top of the file just below # aws-api header. During the release, the version updater will replace the DEV with the proper release number and release date.
• (Optional) If releasing a beta release, create a file named VERSION_SUFFIX at the root of the project, containing a beta release suffix such as -beta01 (with hyphen).

post-release checklist

• When successful, the released artifacts will appear in Maven Central possibly after a few minutes delay.
• Make release announcements.
• (Optional) close any open issues that are fixed by the release.

release announcements

An aws-api release should be announced in the following places:

• Clojure google group
• Clojurians Slack, #releases, #announcements, and #aws channels
  • Post first in #releases
  • Cross-post to the other two channels (linking to the #releases post)

To craft the release verbiage, look at previous announcements, or use the following example. Include hyperlinks of the issue numbers to the GitHub issues.

Subject: [ANN] Nubank's aws-api 0.8.666

Body:

Nubank's aws-api 0.8.666 is now available!

0.8.666 / 2023-04-27
  * 301 gets cognitect.anomalies/incorrect instead of cognitect.anomalies/fault #237
0.8.664 / 2023-04-26
  * Safely return byte arrays from ByteBuffers, honoring the position and remaining attributes, copying the underlying byte array when necessary. #238
  * Upgrade to com.cognitect/http-client "1.0.123", which includes status and headers in anomalies.
    * Fixes #171
    * Improves #15

Obs: the anomaly fix provides users the opportunity to detect 301s and programmatically recover when the "x-amz-bucket-region" header is present.

README: https://github.com/cognitect-labs/aws-api/
API: https://cognitect-labs.github.io/aws-api/
Changelog: https://github.com/cognitect-labs/aws-api/blob/master/CHANGES.md
Upgrade Notes: https://github.com/cognitect-labs/aws-api/blob/master/UPGRADE.md
Latest Releases of api, endpoints, and all services: https://github.com/cognitect-labs/aws-api/blob/master/latest-releases.edn

Note: there should be a public GitHub issue for any significant fix or change. Create a public issue if necessary.

implementation details

A few words about some lower-level implementation details which the release script makes use of.

• The build/revision script calculates the release version (see doc/versioning.md for more details).
• The version_updater.clj tool is responsible for updating four files with the release version: README.md, CHANGES.md, UPGRADE.md, and latest-releases.edn. It makes use of the build/revision script. Note the deps.edn alias update-versions which the release script uses to invoke this tool.

REPL

There are various options for opening a REPL.

cider-nrepl

1. M-x cider-jack-in

CIDER jack-in is a convenient way to start up an nrepl without having to manually configure the nrepl and cider-nrepl dependencies. However, if you wish to use additional deps.edn aliases (e.g. dev, examples), you will have to tweak CIDER's default jack-in command. There are a couple ways to do this:

• Set cider-clojure-cli-global-options to e.g. "-M:dev" (global setting for all projects).
• Set cider-clojure-cli-aliases to e.g. ":dev:examples"
• Set either of the two aforementioned settings locally via .dir-locals.el, e.g.

((nil . ((cider-clojure-cli-aliases . ":dev:examples"))))

• Prefix (C-u) the M-x cider-jack-in command, which allows you to edit the default jack-in command before it runs.

2. M-x cider-connect

• Manually start an nrepl via CLI, then connect to it using CIDER. One option is to define a cider/nrepl alias. Add to e.g. ~/.clojure/deps.edn:

 :aliases {:cider/nrepl {:extra-deps {nrepl/nrepl {:mvn/version "x.y.z"}
                                      cider/cider-nrepl {:mvn/version "a.b.c"}}
                         :main-opts ["-m" "nrepl.cmdline" "--middleware" "[cider.nrepl/cider-middleware]"]}}

• From the project root, start nrepl

clj -M:dev:cider/nrepl

• Connect to nrepl from CIDER (C-c M-c) using indicated host and port.

IntelliJ IDEA / Cursive

1. Configure a REPL Run Configuration ("Run"->"Edit Configurations").
2. Click plus "+" to "Add new Configuration".
3. Select "Clojure REPL, Local".
4. In the configuration window

• Give it a display name
• Select "nREPL" for type of REPL to run
• Select "Run with Deps" for "How to run it". Add any additional desired aliases as a comma-delimited string, e.g. dev:examples.

Now the configured REPL can be launched (run or debug) via "Run" top-level menu or Run Configurations dropdown.

Implementation Notes

http client

This library makes uses of an underlying http client object to aid in http request/response. Originally that client has been the com.cognitect/http-client library. More recently, we've developed a newer client based on the java.net.http package which is present in JDK 11 and up.

What follows is notes on http clients.

timeouts

In the context of http clients and connections, "timeout" may refer to one of (at least) three values:

• name resolution timeout aka resolve timeout
• connect timeout
• read response timeout aka idle timeout

For the cognitect client (which is based on an underlying Jetty http client):

• resolve timeout defaults to 5 seconds
• connect timeout defaults to 5 seconds
• idle timeout is unspecified, which means unbounded

For the java.net.http client, there is no distinct "resolve timeout", therefore we have set the "connect timeout" to be 10 seconds, encompassing the sum of the previous client's resolve and connect timeout defaults.

• resolve timeout (not applicable)
• connect timeout defaults to 10 seconds
• idle timeout is unspecified, which means unbounded

restricted headers

Potentially confusing because "restricted headers" may be referring to one of two separate topics:

• The headers which are disallowed by the java.net.HttpClient implementation, because the implementation will provide its own values for these headers (doc).
• The headers which are forbidden by the Fetch API spec which runs in browsers (doc)

aws-api is concerned with only the first topic.