Skip to main content

Save and Restore Cache from S3

Caching has two primary benefits:

  • Run pipelines faster by reusing the expensive fetch operation data from previous builds.
  • Share data across stages.

You can cache data to an AWS S3 bucket in one stage using the Save Cache to S3 step, and restore it in the same stage or a following stage using the Restore Cache From S3 step. You cannot share access credentials or other Text Secrets across stages.

This topic explains how to configure the Save Cache to S3 and Restore Cache From S3 steps in Harness CI.

info

If you are using Harness Cloud build infrastructure, you can use Cache Intelligence to automate caching.

S3 bucket and AWS connector requirements

You need:

  • A dedicated S3 bucket for your Harness CI cache operations.
  • An AWS connector with read/write access to your S3 bucket.
  • An optional lifecycle configuration, if you want to automatically delete old cache data from your S3 bucket.

Here is an example of an S3 cache bucket policy:

{
"Version": "2012-10-17",
"Statement": [
{
"Sid": "AllowS3BucketAccess",
"Effect": "Allow",
"Action": [
"s3:PutObject",
"s3:GetObject",
"s3:ListBucket",
"s3:DeleteObject"
],
"Resource": [
"arn:aws:s3:::your-s3-bucket/*",
"arn:aws:s3:::your-s3-bucket"
]
},
{
"Sid": "AllowDescribeRegions",
"Effect": "Allow",
"Action": "ec2:DescribeRegions",
"Resource": "*"
}
]
}
warning

You must use a dedicated bucket for your Harness CI cache operations. Do not save files to the bucket manually. The Restore Cache operation will fail if the bucket includes any files that do not have a Harness cache key.

Caching with non-private ACL

If your bucket's ACL is set to something other than private (such as blank, bucket-owner-full-control, or something else), then you must add a stage variable named PLUGIN_ACL. Set the value to the relevant ACL value.

The default value for PLUGIN_ACL is private, so if your bucket's ACL is something other than private, you must set this environment variable accordingly.

Add the Save Cache to S3 step

Add the Save Cache to S3 step after steps that build and test your code, as shown in this diagram:

Here is a YAML example of a Save Cache to S3 step.

              - step:
type: SaveCacheS3
name: Save Cache to S3
identifier: Save_Cache_to_S3
spec:
connectorRef: AWS_Connector
region: us-east-1
bucket: your-s3-bucket
key: cache-{{ checksum filePath1 }} # example cache key based on file checksum
sourcePaths:
- directory1 # example first directory to cache
- directory2 # example second directory to cache
archiveFormat: Tar
...

Save Cache to S3 step settings

The Save Cache to S3 step has the following settings. Depending on the stage's build infrastructure, some settings might be unavailable or optional. Settings specific to containers, such as Set Container Resources, are not applicable when using the step in a stage with VM or Harness Cloud build infrastructure.

Name

Enter a name summarizing the step's purpose. Harness automatically assigns an Id (Entity Identifier Reference) based on the Name. You can change the Id.

AWS Connector

The Harness AWS connector to use when saving the cache to S3. The AWS IAM roles and policies associated with the account used in the Harness AWS Connector must be able to write to S3.

info

This step supports AWS connectors using AWS access key, Assume IAM role on delegate, and IRSA authentication methods without cross-account access (ARN/STS).

IRSA requires modifications to the delegate YAML and build infrastructure settings. The service account to be used for uploading to S3 must be specified in the delegate YAML and in the Service Account Name in the stage's build infrastructure settings.

This step doesn't support AWS connectors that have enabled cross-account access (ARN/STS) for any authentication method.

For more information about roles, permissions, and configuration for AWS connectors, go to:

Region

Define the AWS region to use when saving the cache, such as the AWS region you selected when you created the AWS S3 bucket. For more information go to the AWS documentation:

Bucket

The AWS S3 bucket where you want to save the cache.

Key

The key to identify the cache.

You can use the checksum macro to create a key based on a file's checksum, for example: myApp-{{ checksum filePath1 }}

With this macro, Harness checks if the key exists and compares the checksum. If the checksum matches, then Harness doesn't save the cache. If the checksum is different, then Harness saves the cache.

The backslash character isn't allowed as part of the checksum value here. This is a limitation of the Go language (golang) template. You must use a forward slash instead.

  • Incorrect format: cache-{{ checksum ".\src\common\myproj.csproj" }
  • Correct format: cache-{{ checksum "./src/common/myproj.csproj" }}

Source Paths

A list of the files/folders to cache. Add each file/folder separately.

Endpoint URL

Endpoint URL for S3-compatible providers. This setting is not needed for AWS.

Archive Format

Select the archive format. The default archive format is Tar.

Override Cache

Select this option if you want to override the cache if a cache with a matching Key already exists.

By default, the Override Cache option is set to false (unselected).

Path Style

If unselected, the step uses Virtual Hosted Style for paths, such as http://bucket.host/key. If selected, the step uses Path Style, such as http://host/bucket/key.

For MinIO, you must use Path Style. Make sure Path Style is true (selected).

By default, Path Style is false (unselected).

Run as User

Specify the user ID to use to run all processes in the pod if running in containers. For more information, go to Set the security context for a pod.

Set Container Resources

Maximum resources limits for the resources used by the container at runtime:

  • Limit Memory: Maximum memory that the container can use. You can express memory as a plain integer or as a fixed-point number with the suffixes G or M. You can also use the power-of-two equivalents, Gi or Mi. Do not include spaces when entering a fixed value. The default is 500Mi.
  • Limit CPU: The maximum number of cores that the container can use. CPU limits are measured in CPU units. Fractional requests are allowed. For example, you can specify one hundred millicpu as 0.1 or 100m. The default is 400m. For more information, go to Resource units in Kubernetes.

Timeout

Set the timeout limit for the step. Once the timeout limit is reached, the step fails and pipeline execution continues. To set skip conditions or failure handling for steps, go to:

Set shared paths for cache locations outside the stage workspace

Steps in the same stage share the same workspace, which is /harness. If your steps need to use data in locations outside the stage workspace, you must specify these as shared paths. This is required if you want to cache directories outside /harness. For example:

  stages:
- stage:
spec:
sharedPaths:
- /example/path # directory outside workspace to share between steps

Add the Restore Cache From S3 step

Add the Restore Cache From S3 step before steps that build and test your code, as shown in the following diagram:

Here is a YAML example of a Restore Cache From S3 step.

              - step:
type: RestoreCacheS3
name: Restore Cache From S3
identifier: Restore_Cache_From_S3
spec:
connectorRef: AWS_Connector
region: us-east-1
bucket: your-s3-bucket
key: cache-{{ checksum filePath1 }} # example cache key based on file checksum
archiveFormat: Tar
...
warning

The key value in this step must match the key value in your Save Cache to S3 step.

Restore Cache from S3 step settings

The Restore Cache from S3 step has the following settings. Depending on the stage's build infrastructure, some settings might be unavailable or optional. Settings specific to containers, such as Set Container Resources, are not applicable when using the step in a stage with VM or Harness Cloud build infrastructure.

Name

Enter a name summarizing the step's purpose. Harness automatically assigns an Id (Entity Identifier Reference) based on the Name. You can change the Id.

AWS Connector

The Harness Connector to use when restoring the cache from AWS S3. If your pipeline also has a Save Cache to S3 step, these steps typically use the same connector.

The AWS IAM roles and policies associated with the account used in the Harness AWS Connector must be able to read from S3.

note

This step supports AWS connectors using AWS Access Key, Assume IAM role on Delegate, and IRSA authentication methods without cross-account access (ARN/STS).

This step doesn't support AWS connectors that have enabled cross-account access (ARN/STS) for any authentication method.

For more information about roles and permissions for AWS connectors, go to:

Region

The relevant AWS region. If your pipeline also has a Save Cache to S3 step, the region is typically the same for both steps.

Bucket

The AWS S3 bucket where the target cache is saved. If your pipeline also has a Save Cache to S3 step, the bucket is typically the same for both steps.

Key

The key identifying the cache that you want to retrieve. If your pipeline also has a Save Cache to S3 step, the key value must be the same to restore the previously-saved cache.

The backslash character isn't allowed as part of the checksum value here. This is a limitation of the Go language (golang) template. You must use a forward slash instead:

  • Incorrect format: cache-{{ checksum ".\src\common\myproj.csproj" }
  • Correct format: cache-{{ checksum "./src/common/myproj.csproj" }}

Endpoint URL

Endpoint URL for S3-compatible providers. This is not needed for AWS.

Archive Format

Select the archive format. The default archive format is Tar.

Path Style

If unselected, the step uses Virtual Hosted Style for paths, such as http://bucket.host/key. If selected, the step uses Path Style, such as http://host/bucket/key.

For MinIO, you must use Path Style. Make sure Path Style is true (selected).

By default, Path Style is false (unselected).

Fail if Key Doesn't Exist

Select this option to fail the step if the specified Key doesn't exist.

By default, this option is set to false (unselected).

Run as User

Specify the user ID to use to run all processes in the pod if running in containers. For more information, go to Set the security context for a pod.

Set Container Resources

Maximum resources limits for the resources used by the container at runtime:

  • Limit Memory: Maximum memory that the container can use. You can express memory as a plain integer or as a fixed-point number with the suffixes G or M. You can also use the power-of-two equivalents, Gi or Mi. Do not include spaces when entering a fixed value. The default is 500Mi.
  • Limit CPU: The maximum number of cores that the container can use. CPU limits are measured in CPU units. Fractional requests are allowed. For example, you can specify one hundred millicpu as 0.1 or 100m. The default is 400m. For more information, go to Resource units in Kubernetes.

Timeout

Set the timeout limit for the step. Once the timeout limit is reached, the step fails and pipeline execution continues. To set skip conditions or failure handling for steps, go to:

Go, Node, and Maven cache key and path requirements

There are specific requirements for cache keys and paths for Go, Node.js, and Maven.

Go pipelines must reference go.sum for spec.key in Save Cache to S3 and Restore Cache From S3 steps, for example:

                  spec:
key: cache-{{ checksum "go.sum" }}

spec.sourcePaths must include /go/pkg/mod and /root/.cache/go-build in the Save Cache to S3 step, for example:

                  spec:
sourcePaths:
- /go/pkg/mod
- /root/.cache/go-build

Caching in multi-stage pipelines

Because each stage runs in an isolated environment, you can use caching to pass data from one stage to the next.

The following diagram illustrates cache usage across two stages.

Caching in parallel or concurrent stages

If you have multiple stages that run in parallel, Save Cache steps might encounter errors when they attempt to save to the same cache location concurrently. To prevent conflicts with saving caches from parallel runs, you need to skip the Save Cache step in all except one of the parallel stages.

This is necessary for any looping strategy that causes stages to run in parallel, either literal parallel stages or matrix/repeat strategies that generate multiple instances of a stage.

To skip the Save Cache step in all except one parallel stage, add the following conditional execution to the Save Cache step(s):

Add the following when definition to the end of your Save Cache step.

              - step:
...
when:
stageStatus: Success ## Execute this step if the stage execution is successful thus far.
condition: <+strategy.iteration> == 0 ## And execute this step if this JEXL condition evaluates to true

This when definition causes the step to run only if both of the following conditions are met:

  • stageStatus: Success: Execute this step if the stage execution is successful thus far.
  • condition: <+strategy.iteration> == 0: Execution this step if the JEXL expression evaluates to true.

The JEXL expression <+strategy.iteration> == 0 references the looping strategy's iteration index value assigned to each stage. The iteration index value is a zero-indexed value appended to a step or stage's identifier when it runs in a looping strategy. Although the stages run concurrently, each concurrent instance has a different index value, starting from 0. By limiting the Save Cache step to run on the 0 stage, it only runs in one of the concurrent instances.

Caching in cloned pipelines

When you clone a pipeline that has Save/Restore Cache steps, cache keys generated by the cloned pipeline use the original pipeline's cache key as a prefix. For example, if the original pipeline's cache key is some-cache-key, the cloned pipeline's cache key is some-cache-key2. This can cause problems if the Restore Cache step in the original pipeline looks for caches with the matching cache key prefix and pulls the caches for both pipelines.

To prevent this issue, Harness can add separators (/) to your GCS cache keys to prevent accidental prefix matching and pulling incorrect caches from cloned pipelines. To enable the separator, add this stage variable: PLUGIN_ENABLE_SEPARATOR: true.

If you don't enable the separator, make sure your cloned pipelines generate unique cache keys to avoid the prefix matching issue.

Troubleshoot caching

Go to the CI Knowledge Base for questions and issues related to caching, data sharing, dependency management, workspaces, shared paths, and more. For example: