feat: support operations on push deployments without an Operator by ab-alien-dev · Pull Request #821 · alienplatform/alien

ab-alien-dev · 2026-10-01T19:15:29Z

Summary

Operations are declared in alien.ts, and push deployments can run them without an Operator.

Declaring operations

Stack API: Stack.operations (OperationsConfig) and the TypeScript .operations({...}) builder. Built-in plugins are keyed by name with their settings and approval rules (an operation name or *; no rule means approval is needed). Published custom plugins go under plugins as name@version or an object. Changing operations takes a new release.
Plugin manifest settings: a plugin declares the settings it accepts (string, secret, or resources, with the env variable it reads). approval is reserved for approval rules and cannot be a setting key.
Sync contract: each OperationsBundleDownload carries the plugin's settings as env (redacted in Debug). SyncRequest.operationsConfig reports an Operator's own declaration without setting values.
Operators installed without a release: read the same JSON from OPERATOR_OPERATIONS (values are literals or { "env": "NAME" }). The Helm chart sets it from remoteOperator.operations, and remoteOperator.operationsEnv adds env entries (for example secretKeyRef) for those values.
CLI: alien operations list --deployment <id> lists a deployment's operations.
Examples: byoc-database declares cloudwatch and aws-vpc; custom-operations declares its settings.

Push deployments without an Operator

Reserved command target: operations-worker is reserved next to operator. Only an operations capability may address it, single-target shorthand never selects it, and its results follow the same rules as operator.
Supplied stacks: DeploymentRecord.supplied_stacks lets a deployment store that receives stacks from its control plane deploy those instead of re-reading the release by id. A same-release change must arrive as UpdatePending. Records without supplied stacks behave as before.
GCP: the Cloud Run worker's push subscription acknowledgement deadline is capped at Pub/Sub's 10–600s range, matching the Terraform emitter.
CLI access requests: the CLI reads approvalChannels from the coordinates and only waits for kubectl approval when that channel applies.

Removing permissions no longer requires setup

Update preflights accept permission profiles, management permissions, and frozen service accounts that only lost grants. The update runs with the installed identities, so it can still delete what the removed grants covered; the next setup trims them. Adding a grant, removing a role guard, or changing a permission set's contents still requires setup. Setup re-import keeps the exact comparison.
A frozen service account that only lost grants keeps its installed definition during the update.
Secret sync leaves a removed secret in place, with a warning, when the cloud denies the delete (403 / access denied), instead of failing the update.

Platform client SDK

The spec copies are regenerated: stack operations, operationsConfig on sync, approvalChannels on access requests, and removal of the per-project operations enablement and policy endpoints.

Release condition

approvalChannels is required on the coordinates response, and operations list --deployment uses a new API route. Do not cut a CLI release from this until the API that serves both is deployed.

Follow-up

operations-worker is reserved as a command target but not rejected as a stack resource id here; the control plane rejects the name at release creation.

Test plan

- Reserve the `operations-worker` command target alongside `operator`: only an operations capability may address it, and single-target shorthand never selects it. - Let a deployment store supply the stacks to deploy (`supplied_stacks`). The deployment loop uses them instead of re-reading the release, so a control plane can add deployment-specific resources. - CLI: read `approvalChannels` from access-request coordinates and only wait for kubectl approval when that channel applies. - Platform client SDK: approval channels and the access-request queue authorization.

greptile-apps · 2026-10-01T19:20:21Z

Confidence Score: 5/5

[High risk] Adds operations support without requiring an Operator component.

The PR appears safe to merge based on this review; no outstanding previous finding or actionable new issue was identified.

Summary

The PR adds stack-declared operations and a reserved operations worker for push deployments, carries control-plane-supplied stacks through deployment updates, and allows grant removals to wait for the next setup. Since the previous review, it also revises acquisition diagnostics, setup checkpoint finalization, and AWS service-account trust and policy reconciliation.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Stack operations declaration] --> B[Release and supplied deployment stack]
  B --> C[Push deployment operations worker]
  C --> D[Reserved operations command target]
  E[CLI setup] --> F[Checkpoint setup revision]
  F --> G[Release manager lease]

_{Reviews (10) · Last reviewed commit: "Merge remote-tracking branch 'origin/mai..." · Reviewed by Greptile}

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json # crates/alien-cli/src/commands/access_requests.rs # crates/alien-manager/src/loops/deployment.rs

- Cap the Cloud Run worker's push subscription acknowledgement deadline at Pub/Sub's 10 to 600 second range, as the Terraform emitter already does. A worker timeout above 600s made the subscription fail to create. - Enable the Pub/Sub API for stacks with a commands-enabled worker, not only for stacks with a queue.

ab-alien-dev · 2026-10-01T19:56:35Z

@greptile-apps review

- Revert enabling the Pub/Sub API for commands-enabled GCP workers. It adds a setup-owned resource, so every installed GCP deployment with such a worker would need a setup rerun on its next update. Commands- enabled workers already relied on the API being on, as it is by default. - Document that a same-release supplied-stack change must arrive as UpdatePending. - Move the misplaced coordinates doc comment to its function and rename it `fetch_approval_instructions`.

ab-alien-dev · 2026-10-01T20:17:49Z

@greptile-apps review

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json # crates/alien-manager/src/traits/mod.rs

Adds Stack.operations_enabled (operationsEnabled in JSON, omitted when false) and the enableOperations() builder method in Rust and TypeScript. Deployments without an Operator get an operations worker and its permission profile only when the stack sets it. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Replaces enableOperations() with operations({...}): built-in plugins by name with settings and approval rules, and published custom plugins pinned by name@version. Plugin manifests declare the settings they accept. An Operator installed without a release reads the same JSON from OPERATOR_OPERATIONS (Helm value remoteOperator.operations), resolves { env: NAME } settings locally, and reports the declaration without values in sync. Bundle targets carry each plugin's settings env, kept in memory only. The Manager forwards the reported declaration. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json # crates/alien-core/src/lib.rs # packages/core/src/generated/schemas/stack.json

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Removing the last vault-backed worker drops the secrets vault, and setup revokes the management role's write access with it. The secret sync then tried to delete the values it wrote earlier and was denied. Forget them instead of deleting them. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

The update keeps setup-owned frozen resources in its target, so the secrets vault stays in the stack after its last worker is removed. Check the management role's vault grant instead of the vault's presence. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

An update that only removes grants runs with the installed identities and leaves trimming them to the next setup. The removed grants are what the update needs to delete the resources and secrets they covered; requiring setup first revoked them, so those deletes were denied. - compatibility checks accept profiles, management permissions and frozen service accounts that only lost grants (dropping a role guard still widens access and still needs setup) - the update keeps a narrowed frozen service account's installed definition - a denied secret delete leaves the value in place instead of failing - a delete denied during an update fails; only teardown counts an unreachable resource as deleted Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…rt-push-mode-model-for-operations # Conflicts: # crates/alien-preflights/src/compatibility/frozen_resources_unchanged.rs

The setup re-import check refuses settings that change the management permissions setup installed. It now compares exactly; only update preflights allow narrowing. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…churn - the settings env on a bundle is stored in the Operator's encrypted state; say so instead of claiming it is never persisted - approval rules match an operation name or *, as implemented - the stack field describes the declaration, not a consumer of it - test that only teardown counts an unreachable resource as deleted - revert formatting-only hunks in unrelated preflight files Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Takes main's executor delete rule (access denied ends a delete except during a replace) and drops this branch's teardown-only variant: with permission narrowing deferred to the next setup, an update still holds the grants it needs to delete what a release removed. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ab-alien-dev · 2026-10-07T14:42:48Z

@greptile-apps review

ab-alien-dev · 2026-10-07T15:17:17Z

@greptile-apps review

ab-alien-dev · 2026-10-07T15:18:26Z

@greptile-apps review

- a service account losing its last resource grant narrows when its profile holds no resource grants either; the uncaptured legacy format still needs setup - plugin manifests may not declare a setting named approval, which operations() uses for approval rules - only 403 or access denied lets a removed secret go; a 401 can be a transient credential failure and still fails the sync - CLI approval text no longer says approval always happens on the customer's side Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ab-alien-dev · 2026-10-07T16:04:00Z

@greptile-apps review

…rt-push-mode-model-for-operations

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json # crates/alien-manager/src/registry_access.rs # crates/alien-manager/src/traits/deployment_store.rs

Stack operations and the operations sync types appear in many schemas, so the filtered client grew to about 240,000 lines, just past the cap. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

lilienblum

One blocker remains: removing an Operator's final locally declared plugin does not report an empty declaration, so the control plane cannot clear its previously reported plugins and approval rules. See the inline finding.

Verdict: request changes.

Coverage: reviewed head a31dd7b against main at 78fa58c, merge base 093d0b5. Inspected the full 105-file change, operations types/builders, sync reporting, reserved command targets and capabilities, supplied-stack selection, permission narrowing, setup re-import strictness, secret cleanup and Helm generation. Ran cargo test -p alien-operations-sdk --lib (75 passed) and cargo test -p alien-preflights --lib compatibility (42 passed). CI still has a worker integration failure and a cancelled Rust job; the failed worker test reports a port-bind error, and I have not attributed it to this change. No live cloud deployment or full manager/deployment test suite was run. The documented API-before-CLI release order still applies.

Removing the last plugin from OPERATOR_OPERATIONS left operationsConfig out of sync, so the platform kept the previous declaration. A current Operator now always reports one, empty when it declares nothing; only an older Operator omits the field. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ab-alien-dev · 2026-10-08T11:38:54Z

@greptile-apps review

Note for this review: the remaining blocker (a denied AWS read during deletion marking a still-existing resource gone) is in the delete-probe flow added on main by #935 (6f44ca0), which reached this branch through a merge. This PR does not change that flow; its only alien-infra changes are the GCP Pub/Sub ack-deadline clamp and two test-support struct fields. Please review it against main rather than attributing it to this change.

ab-alien-dev · 2026-10-08T11:40:19Z

@greptile-apps review

The remaining blocker (a denied AWS read during deletion clearing a resource that may still exist) is in the delete-probe flow added on main by #935 and reached this branch through a merge; this PR does not change that code. It is tracked separately in #949 to be fixed against main.

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json

ab-alien-dev · 2026-10-08T12:07:32Z

@greptile-apps review

lilienblum

…rt-push-mode-model-for-operations # Conflicts: # crates/alien-manager/src/traits/mod.rs

ab-alien-dev · 2026-10-08T15:12:59Z

@greptile-apps review

…ut the plugins key (ALIEN-1358) (#968) ## What changes Since #821, `OPERATOR_OPERATIONS` (the `remoteOperator.operations` Helm value) uses the `OperationsConfig` shape: ```json {"plugins": {"kubernetes": {"approval": {"get-pods": "auto"}}}} ``` Before #821, the value was the plugin map itself: ```json {"kubernetes": {"approval": {"get-pods": "auto"}}} ``` Values written in the earlier shape still deserialize, but serde ignores the unknown top-level key. The Operator then starts with an empty declaration and runs no operations. Nothing reports the problem: the Operator logs no error, and it never requests an operations bundle. An installation upgraded to an Operator built after #821 silently loses every operation. `parse_operator_operations` now reads a non-empty JSON object that has neither a `plugins` nor a `custom` key as the plugin map, and logs a warning naming the shape. Objects in the current shape, and `{}`, parse exactly as before. Setting validation is unchanged. The format probe only chooses whether to add the `plugins` wrapper. Typed deserialization uses the original JSON text, so duplicate known fields remain errors in both formats. ## What the next maintainer inherits `with_plugins_key` in `alien-operator/src/cli.rs` is the only place the earlier shape is accepted. It can go once no installation sets `remoteOperator.operations` without `plugins`. ## Verification - New test `operator_operations_read_a_plugin_map_without_the_plugins_key`. It fails on main, where the earlier shape yields no config, and passes with this change. It also checks that both shapes parse to the same config. - Original-head validation: `cargo test -p alien-operator --lib operator_operations`: 2 passed. - `operator_operations_reject_duplicate_fields` rejects duplicate `plugins`, `custom`, `approval`, `settings`, custom plugin name, and custom plugin version fields. It fails on the original PR head and passes with the fix. - An isolated harness compiled the exact production parser, configuration types, error types, and parser tests with lockfile dependency versions: 4 tests passed, including all eight duplicate-field cases. Formatting and diff checks passed.

greptile-apps Bot reviewed Oct 1, 2026

View reviewed changes

Comment thread crates/alien-manager/src/loops/deployment.rs

This comment has been minimized.

Sign in to view

ab-alien-dev added 2 commits October 1, 2026 22:34

greptile-apps Bot reviewed Oct 1, 2026

View reviewed changes

Comment thread crates/alien-preflights/src/mutations/gcp_service_activation.rs Outdated

ab-alien-dev and others added 16 commits October 3, 2026 19:32

Merge remote-tracking branch 'origin/main' into alan/alien-1065-suppo…

d483f76

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json # crates/alien-manager/src/traits/mod.rs

chore: enable operations in the byoc-database example

b9af6b9

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore: refresh platform SDK spec with stack operations

7bb2e37

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chore: refresh platform SDK spec after merging main

8bcf698

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

fix: add plugin settings to the custom-operations example manifest

f2c64cc

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Merge remote-tracking branch 'origin/main' into alan/alien-1065-suppo…

34daf4a

…rt-push-mode-model-for-operations # Conflicts: # crates/alien-preflights/src/compatibility/frozen_resources_unchanged.rs

chore: regenerate core schemas for operations doc changes

a984c66

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

greptile-apps Bot reviewed Oct 7, 2026

View reviewed changes

Comment thread packages/core/src/operations.ts

ab-alien-dev and others added 3 commits October 7, 2026 19:20

Merge remote-tracking branch 'origin/main' into alan/alien-1065-suppo…

be9e63b

…rt-push-mode-model-for-operations

chore: raise the filtered Platform client size cap

a31dd7b

Stack operations and the operations sync types appear in many schemas, so the filtered client grew to about 240,000 lines, just past the cap. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

lilienblum requested changes Oct 8, 2026

View reviewed changes

Comment thread crates/alien-operator/src/loops/sync.rs Outdated

ab-alien-dev and others added 2 commits October 8, 2026 14:20

fix: import the reported operations helper in operator sync tests

74f6031

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ab-alien-dev mentioned this pull request Oct 8, 2026

AWS delete: a denied existence check can forget a resource that still exists #949

Open

ab-alien-dev requested a review from lilienblum October 8, 2026 11:47

Merge remote-tracking branch 'origin/main' into alan/alien-1065-suppo…

ab2c44b

…rt-push-mode-model-for-operations # Conflicts: # client-sdks/platform/openapi.json # client-sdks/platform/rust/openapi-3.0.json # client-sdks/platform/rust/openapi.json

lilienblum approved these changes Oct 8, 2026

View reviewed changes

Merge remote-tracking branch 'origin/main' into alan/alien-1065-suppo…

9311fa9

…rt-push-mode-model-for-operations # Conflicts: # crates/alien-manager/src/traits/mod.rs

ab-alien-dev merged commit d860dd1 into main Oct 8, 2026
24 checks passed

ab-alien-dev deleted the alan/alien-1065-support-push-mode-model-for-operations branch October 8, 2026 15:32

lilienblum mentioned this pull request Oct 9, 2026

fix(operator): read OPERATOR_OPERATIONS written as a plugin map without the plugins key (ALIEN-1358) #968

Merged

Conversation

ab-alien-dev commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Declaring operations

Push deployments without an Operator

Removing permissions no longer requires setup

Platform client SDK

Release condition

Follow-up

Test plan

Uh oh!

greptile-apps Bot commented Oct 1, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Confidence Score: 5/5

Summary

Diagram

Uh oh!

Uh oh!

This comment has been minimized.

ab-alien-dev commented Oct 1, 2026

Uh oh!

Uh oh!

ab-alien-dev commented Oct 1, 2026

Uh oh!

ab-alien-dev commented Oct 7, 2026

Uh oh!

ab-alien-dev commented Oct 7, 2026

Uh oh!

ab-alien-dev commented Oct 7, 2026

Uh oh!

Uh oh!

ab-alien-dev commented Oct 7, 2026

Uh oh!

lilienblum left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

ab-alien-dev commented Oct 8, 2026

Uh oh!

ab-alien-dev commented Oct 8, 2026

Uh oh!

ab-alien-dev commented Oct 8, 2026

Uh oh!

lilienblum left a comment

Choose a reason for hiding this comment

Uh oh!

ab-alien-dev commented Oct 8, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ab-alien-dev commented Oct 1, 2026 •

edited

Loading

greptile-apps Bot commented Oct 1, 2026 •

edited

Loading