Repository navigation
feat(build): build and release a sandbox base image like every other compute type - #700
Merged
ItamarZand88 merged 7 commits intoSep 24, 2026
Conversation
|
Contributor
Author
|
@greptile review |
…te type A sandbox whose image is a local artifact directory is pushed, but never recorded in the push cache, never matched against it, and never rebased under `--prebuilt`. So an unchanged sandbox is re-pushed on every release, and a prebuilt release from another checkout carries a path that no longer exists. The three sites that make a push cheap enumerated the worker, the container and the daemon. The sandbox now joins them, reading the same local-directory test the push path already applies. A bundle URI and a catalog name reach these sites too and name no artifact, so they pass through untouched. The note that tells the next person where to add a compute type pointed at two places and left these three out, which is how the sandbox came to be handled on one side of the release and not the other.
A sandbox was the only compute resource `alien build` ignored. Declaring one from source meant a refusal at plan time, so the base image had to be built and pushed by hand while every other resource was built by the CLI. `alien build` now collects a sandbox declared from source and produces an OCI artifact directory for it, and the release pushes that directory the way it already pushed a hand-built one. The bundle still layers the agent on top, and AWS still builds the MicroVM from it inside the customer's account. Source is refused only where nothing can build it. On AWS it is a declaration; elsewhere the image is pulled as written, so an unbuilt source would schedule a workload with no image and the refusal names the platform. The base image is a root filesystem rather than a binary on a base Alien picks, so it is built from a Dockerfile and the other toolchains are refused by name.
…e means A partitioned CI build writes one artifact directory per runner group and merges them afterwards. The merge read and wrote the image of a worker, a container and a daemon, so a sandbox built on one runner was dropped on the way out. The planner needs no arm. It asks which resources produce a host binary, and a sandbox base image is built from a Dockerfile, so it never does. That is now written where someone would otherwise go looking for the missing branch. The declaration's own documentation said source was built at deploy time, which was never true and is now wrong in a second way.
Four tests render a Live sandbox and three render a remote one, and none render both, so the grant a remotely published Live sandbox depends on was asserted nowhere.
The refusal said a Live sandbox's image is built after setup, so setup cannot bind to it. That is the diagnosis and it stopped there, leaving a developer who hit it with nothing to try. It now names the two ways through: add the sandbox with `remoteAccess` and reach it through a remote binding, or give it a base image that needs no build so it can stay Frozen and be linked. A sandbox built from source is always the first case, because its image is private and a private base image is Live only.
…rough the runner The Live grant test asserted with `contains`, which is monotone: it passes just as happily on a widened action list or a wildcarded image ARN, and it checked four of the seven actions. It now carries the snapshot its Frozen sibling already had, asserts the policy lands on the Remote Bindings identity, and names every action. The snapshot's grant is byte-identical to the Frozen one, which is what the test claims and could not previously show. The per-platform source gate was covered by calling `validate_for_platform` directly. That check once sat unregistered and method-level tests missed it, which is why this file exists at all, and the Kubernetes runtime's empty-image fallback rests on the gate being registered rather than on the method refusing. The new case runs the registry for both Kubernetes and AWS. A comment in that fallback justified itself with a guarantee this branch removes, that Source is refused on every platform. The permission set's header made a claim a Live sandbox on AWS breaks, since its image lands after the stack.
A source build is pushed to the project's own repository, so it is private by the time anything reads the field that says so. The refusal keyed on that field, which meant a Frozen source sandbox passed the build-time gate and failed later, when setup tried to pull an image the registry had not opened to an account the deployment had not yet reported. It is refused on the declaration instead, where the developer is.
ItamarZand88
force-pushed
the
itamar/alien-941-make-the-sandbox-image-follow-the-worker-build-and-release
branch
from
September 24, 2026 17:45
14e96b0 to
c430261
Compare
Contributor
Author
|
@greptileai review |
Contributor
Author
|
@greptile review |
ItamarZand88
deleted the
itamar/alien-941-make-the-sandbox-image-follow-the-worker-build-and-release
branch
September 24, 2026 18:39
ItamarZand88
added a commit
that referenced
this pull request
Sep 25, 2026
## Summary AWS sandboxes can now be deployed with `alien-deploy`, the direct path with no Terraform or CloudFormation. Setup creates everything the runtime identity is never allowed to create itself, and the runtime builds only what it is granted. Step-by-step flow when `alien-deploy deploy` sets up a stack with a sandbox: 1. **Setup, using the deployer's credentials, creates the sandbox's build role. For `egress: deny` it also creates an operator role, a loopback-only security group and a network connector. It records each object and passes them to the sandbox's controller.** ← the heart 2. A Frozen sandbox's image is built during setup and rolled once when setup runs again with a new bundle. A Live sandbox's image is built later by the runtime, which passes the setup-made role. 3. Destroy removes the images, then everything setup created, in dependency order. This PR changes AWS sandboxes from template-only to deployable directly, with the same objects the templates create. ## What was broken `alien-deploy` could not deploy an AWS sandbox. The build role, the deny egress objects and the remote-access grant exist only in the Terraform and CloudFormation setups. The runtime identity is deliberately never given `iam:CreateRole`, so it cannot make them. ## What I did - **Setup scaffolding.** A setup step creates the build role, and for deny egress an operator role, a loopback-only security group and a network connector (through Cloud Control). It also writes a published sandbox's remote-access grant onto the Remote Bindings role. Each object is tagged, recorded in the deployment's runtime metadata and verified before it is adopted. - **Frozen sandboxes on the direct path.** Setup builds the image. A setup re-run with a new bundle re-grants the role and rolls the image once. The runtime never builds or deletes a Frozen image. - **Guards.** Two Deny statements on the management identity block it from writing to setup's sandbox roles and to its own role. The drift check lets an update add either guard, since a Deny only narrows the identity, but not remove one. An earlier version must accept stacks carrying them, and a separate PR (#718) adds that tolerance on main. It has to land first. - **Updates that need setup are refused.** A runtime update that would change scaffolding (a new sandbox, a different egress, network or build policy, or publishing and unpublishing) is refused with the reason, until setup runs again. - **Teardown.** It deletes the image, then the connector, the security group (after the connector's network interfaces are released), the roles and the grant. Anything already gone is skipped. - **Destroy after a failed setup.** When setup fails before the management identity exists, and nothing runtime cleanup acts on has come up, the manager skips runtime cleanup. The destroy goes straight to setup teardown. Before this, it ended in `delete-failed`, and the build role stayed in the account. ## Files touched - `crates/alien-infra/src/setup_scaffolding/`: the scaffolding step, AWS sandbox roles and egress, seeding and teardown. - `crates/alien-infra/src/sandbox/aws.rs`: Frozen build, roll and ownership rules on the direct path. - `crates/alien-deployment/src/`: runs the scaffolding in initial setup, refuses updates that need setup, and tears scaffolding down. - `crates/alien-core/src/`: the runtime-metadata record and the build role naming. - `crates/alien-permissions/permission-sets/`: the two management guards. - `crates/alien-preflights/src/`: the build role name budget and the guard exemption. - `crates/alien-aws-clients/src/`: the Cloud Control, EC2 and IAM calls the scaffolding needs. - CloudFormation and Terraform generator snapshots, updated for the guards. ## How I tested - **Manually (us-west-2):** did a direct deployment with the local control plane, using a Live and a Frozen sandbox with open egress. - Setup created both build roles and built the Frozen image under setup tags. - A remote-access session on the Live sandbox started in 2.4s and ran a probe. - Setup re-run with a new Frozen bundle rolled that image exactly once. - Added Live and Frozen `deny` sandboxes. Setup created their operator roles, loopback-only groups and ACTIVE connectors. Sessions started and TCP out was blocked. - `alien-deploy destroy` removed every object setup and the runtime created. The account inventory matched its pre-run baseline. - With a standalone manager, `alien deploy` failed in setup after creating the build role. `alien destroy` then reached `deleted` and removed the role. - **Unit tests:** `alien-infra` 741, `alien-deployment` 141, `alien-manager` 263, plus preflights and core. Tests use strict mocks for every IAM, EC2, Cloud Control and MicroVM call, and the behavioural fixes are mutation-tested. - **Anything I couldn't test:** - Deny sessions were started with admin credentials; the compute grant comes from `sandbox/management` and was checked by configuration only. - Under `egress: deny`, DNS still resolves, because the security group doesn't cover the VPC resolver. That gap also exists in the template setups and is tracked separately. I also ran a security review on the diff. What it checked: - **The runtime minting a role:** `sandbox/provision` grants `iam:PassRole` on the setup-made role only and never `iam:CreateRole`. The management guards deny writes to setup's sandbox roles and to the management role (`aws_management_role_guards` snapshot). - **Adopting a role someone else made:** a same-named role is adopted only if it carries setup's tags and its trust, inline and attached policies match exactly. Otherwise setup refuses (`aws_sandbox.rs` adoption tests). - **Leaving a remote grant behind:** unpublishing requires setup, and setup revokes the grant even when the stack no longer declares the Remote Bindings role. - **A destroy that skips cleanup it owes:** the rule returns nothing once a Live resource, a synced secrets vault or a Frozen type with runtime cleanup has come up. It also returns nothing on any resource without a lifecycle (`deleting.rs` tests). - **Deny egress holes:** the group has no ingress and egress only to `127.0.0.1/32`, and a tagged group that allows more is repaired or refused (`aws_sandbox_egress.rs` tests). Nothing turned up. Depends on #700 (base), #715 (image tagging) and the #712 → #713 → #714 chain (image read, build inputs, teardown). This branch merges #714 and #715 and will rebase once they land.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
A sandbox was the only compute resource
alien buildignored, so a developer whowanted one on their own image built and pushed it by hand while the CLI built
everything else. It now goes through the same two commands as a Worker.
When a developer runs the flow, three things happen:
alien buildsees a sandbox declared from source and builds its base imagefrom a Dockerfile, writing the same kind of artifact directory a Worker gets.
alien releasepushes that directory and stores a location-free reference —the heart of this change, and the step that means nobody types a registry path.
from it inside the customer's account, exactly as before.
This turns a hand-run
docker pushintoalien buildandalien release.What was broken
Declaring a sandbox from source was refused at plan time on every platform, so
the base image had to exist before the stack did. Three places in the release
path knew about a worker, a container and a daemon but not a sandbox, so an
unchanged sandbox was pushed again on every release and
--prebuiltcarried apath that no longer existed.
What I did
A sandbox base image is a root filesystem rather than a binary laid on a base
Alien picks, so it is built from a Dockerfile and the other toolchains are
refused by name. Source is now legal only where something can build it: on AWS
it is a declaration, and elsewhere the refusal names the platform, because there
the image is pulled as written and an unbuilt source would schedule a workload
with no image.
Two things a reviewer might look for and not find. The planner needs no sandbox
arm, because it asks which resources produce a host binary and a Dockerfile
build never does; that reasoning is now a comment where the missing branch would
be. And the preflight that refuses a resource linking a Live sandbox now names
the two ways out instead of stopping at the diagnosis.
Files touched
crates/alien-build/src/lib.rs— the sandbox build branch, and the Docker-onlytoolchain policy.
crates/alien-build/src/{merge,plan,toolchain/mod}.rs— a sandbox survives thepartitioned CI merge; the new workload kind.
crates/alien-cli/src/commands/release.rs— the three push-cache and rebasesites.
crates/alien-core/src/resources/sandbox.rs— the per-platform source gate.crates/alien-preflights/— the refusal message, and a test that drives thegate through the runner rather than calling the method.
crates/alien-cloudformation/tests/— the Live remote grant, now snapshotted.How I tested
Ran the whole flow against real AWS rather than only in tests.
alien build --platforms awson a stack declaring a sandbox from source:produced
.alien/build/aws/alien-sandbox-b492f56d/linux-aarch64.oci.tar.alien release --platforms aws: pushed it and stored the proxy reference. Asecond release did not push again, which is the cache fix working.
FROM <account>.dkr.ecr.<region>.amazonaws.com/<project repository>:<tag>,so the image the developer built is the one the MicroVM is built on, and that
the bundle landed in every regional store.
cargo testacross the touched crates: 1054 pass. Two pre-existing dockerpull/export failures reproduce on a clean
main.This PR touches a grant and a permission set, so I walked the boundary:
project from the deployment record rather than the stack and refuses anything
else, pinned by
a_refused_sandbox_base_image_leaves_the_grant_incomplete.microvm-image:${AWS::StackName}-agentsARN with no trailing wildcard, so asibling sandbox in the same stack is not matched. Its policy block is
byte-identical to the Frozen one.
Bindings role alone.