What's Changed

• Add remote_cache: nx, Nx's self-hosted remote cache on the RunsOn agent by @crohr in #86

Full Changelog: v2.7.0...v2.7.1

What's Changed

• Add remote_cache: turbo, Turborepo's remote cache on the RunsOn agent by @crohr in #85

Full Changelog: v2.6.0...v2.7.0

What's Changed

• Build the action in CI and release the build CI tested by @crohr in #74
• Skip post-step work after an EC2 Spot interruption by @crohr in #69
• Add sticky_save input for restore-only sticky caches by @crohr in #73
• Rename internal/cache to internal/magiccache by @crohr in #76
• Run CI on every branch by @crohr in #77
• Add a gocacheprog input that shares Go's build cache through Magic Cache by @crohr in #75
• Add a bazel_remote_cache input that gives Bazel a remote cache through Magic Cache by @crohr in #79
• Add a remote_cache input: Bazel, Go, Gradle, mbx and sccache caches on the RunsOn agent by @crohr in #81
• Build with Go 1.27.3 to block the compromised eMudhra intermediates by @crohr in #84

Full Changelog: v2.5.0...v2.6.0

What's Changed

New

• Disk metrics report free and total bytes (disk_free, disk_total), and the post-step summary charts them.
• Disk metrics identify each mount's EBS volume with a new VolumeId dimension, so a sticky disk or snapshot volume on /var/lib/docker reports its own volume instead of the root volume's. (#68, closes #19)

Changed

• On RunsOn v3.4.0 and later, the runner agent reports the job cost in the "Complete runner" step, and show_costs only chooses how it is shown (inline, summary or off). On those stacks the action no longer calls the RunsOn cost API or DescribeAvailabilityZones. Older stacks keep the post-step cost report. (#70)
• Disk metrics now carry device and VolumeId dimensions. CloudWatch matches exact dimension sets, so dashboards or alarms that select disk metrics by their dimensions need the new ones.

What's Changed

New

• sticky_cache: tool-cache persists RUNNER_TOOL_CACHE on the sticky disk, so toolchains downloaded by setup-* actions are reused by later jobs in the same sticky lineage. Linux and Windows. Run it before the setup-* actions. The mode hides the runner image's preinstalled toolchains for the job; read the README section before enabling it. (#54, #55)

Changed

• sccache: s3 scopes its S3 key prefix per repository and platform: cache/sccache/<repository-id>/<os>-<arch>/v1 instead of the shared cache/sccache. The first run after upgrading starts a cold cache for each repository and platform. Objects under the old prefix expire through the stack's cache lifecycle rule. Thanks @garysassano (#58)
• Cost reporting is skipped when pricing is unavailable: the action logs an informational message instead of a warning, and prints no cost table or job summary. (#61)

Fixed

• sticky_cache: git: sparse and filter: blob:none checkouts of large repositories no longer fail with HTTP 400 / expected 'packfile' when git fetches the missing blobs. Checking the requested objects against the mirror is also much faster (one pass instead of one git process per object). Thanks @zackerydev (#64)

Full Changelog: v2.3.1...v2.4.0

What's Changed

• Enable sccache on Windows runners by @crohr in #56
• Continue cold when sticky disks are unavailable by @crohr in #59

Full Changelog: v2.3.0...v2.3.1

What's Changed

• Remove global sudo from action wrapper by @crohr in #42
• Add sticky disk cache modes by @crohr in #43
• Clean interrupted Git repacks before sticky snapshots by @crohr in #48
• Scope cold Git mirrors to workflow history by @crohr in #50
• Fix warm cache symlink merges by @crohr in #52
• Authenticate read-only Git LFS lock requests by @crohr in #51
• Refactor sticky cache runtime state by @crohr in #46
• Allow verified Windows sticky mount roots by @crohr in #53

Full Changelog: v2.2.0...v2.3.0

Full Changelog: v2.1.2...v2.2.0

Full Changelog: v2.1.1...v2.1.2

What's Changed

• Upgrade go, deps, and sanitize values better by @crohr in #30
• Propagate actions runtime token by @crohr in #34

Full Changelog: v2.1.0...v2.1.1