Persistent disks

A persistent disk keeps files between jobs of one repository. Use it for dependency stores, build outputs, toolchains, or anything else that is slow to recreate and safe to rebuild.

Persistent disks are enabled per organization. If your repository settings do not show Persistent disk, contact support@avrea.com. They are available only in datacenters with persistent storage.

Persistent disks are a paid feature: see Pricing.

Enable a persistent disk

1. Open the repository's Settings in the Avrea console.
2. Under Persistent disk, turn on the operating systems that should get a disk. Linux, macOS and Windows each get their own disk and filesystem.
3. Optionally set the size: 10 to 500 GB, 50 GB by default. Size can only grow; a larger size applies the next time a disk is attached.

Workflows find the disk through $PERSISTENT_DISK_PATH:

Operating system	Path
Linux	/mnt/persistent
macOS	/Volumes/AVREA_PERSISTENT
Windows	P:\

workflow.yml

- name: Restore dependencies from the persistent disk
  run: |
    mkdir -p "$PERSISTENT_DISK_PATH/deps"
    ./scripts/install-deps --cache-dir "$PERSISTENT_DISK_PATH/deps"

Persistence is best effort. When storage is unavailable, the job runs without the disk, so workflows should also work from an empty or missing disk.

How branches share the disk

Each job gets its own copy of the disk. Whether its changes are kept depends on the branch:

• Writer branches publish. By default only the repository's default branch (for example main) is a writer. When a writer job finishes, its copy becomes the new disk, so the most recently published default-branch run is what later jobs start from. Change the list under Writer branches; use exact branch names, and $default for the default branch.
• Pull requests and other branches start from the published disk. They receive a writable copy of the latest published contents. Their changes are discarded when the job ends and never reach the published disk or other jobs.
• Concurrent writers do not merge. Writer jobs that run at the same time start from the same published version. The first one to finish publishes; the others started from an older version, so their changes are discarded. An older copy never overwrites newer contents.
• Only cleanly finished jobs publish. A job whose steps failed still publishes if the runner finished normally. Cancelled and crashed jobs do not publish.

Treat the disk as readable by any job

Every job in the repository, including pull requests from forks, can read the published disk. Store only content you can rebuild, and never store secrets or credentials on it.

Docker and Git caches

Docker cache on persistent disk and Git cache on persistent disk store their data on the Linux disk and follow the same branch rules. They do not add a separate storage charge.

Reset or disable

• Reset erases one operating system's disk and provisions a fresh one. Jobs may wait for the replacement or run without the disk meanwhile.
• Disable permanently deletes every persistent disk of the repository, in all operating systems and datacenters, including Docker and Git caches. Re-enabling creates fresh, empty disks.

Disks that are in use must finish their current activity before a reset or disable is accepted. Neither action can be undone.

Pricing

Persistent disks cost $0.50 per provisioned GB per month, prorated for partial months.

• Charges are for provisioned size, not used space. Idle disks are billed while they exist.
• Each operating system and datacenter disk is billed separately. One 50 GB disk costs $25 for a full month; Linux and macOS disks of 50 GB each cost $50.
• Charges start when a disk is ready and stop as soon as a disable is accepted. A reset pauses charges until the replacement is ready.
• Per-job copies are not billed separately.

Persistent-disk charges draw on the same monthly credit as runner minutes. See Billing.