feat(BA-6581): wire IdleCheckerRepository into idle-check reconciler source by seedspirit · Pull Request #12398 · lablup/backend.ai

seedspirit · 2026-06-24T09:06:19Z

Summary

Add IdleCheckerRepository (with its DB source and result types) that fetches a normalized IdleCheckSnapshot of sessions, their applicable scopes, scope→checker bindings, and typed checker specs.
Wire the repository through the orchestration DI (composer / sokovan input) and the reconciler stage factory so IdleCheckSource fetches real data instead of returning a placeholder.
Model the checker spec as a single IdleCheckerSpec (a type discriminator plus optional per-kind sub-specs session_lifetime / network / utilization), stored in the idle_checkers.spec JSONB column via the existing PydanticColumn. A model_validator enforces that exactly the matching sub-spec is present for the chosen type. This replaces the previous ABCColumn + hand-written serialize/load dispatch. Concrete spec fields land with the checker-logic stories.
Move the spec and its CheckerType discriminator out of the ORM layer to common/data/idle_checker/types.py (pure Pydantic, ORM-agnostic). Manager-internal projections (ScopeType, ScopeRef, IdleCheckSessionView) stay in manager/data.

No DB migration: both ABCColumn and PydanticColumn map to JSONB and the idle_checkers table is unreleased.

Incidental test-infra fix

test: eagerly call ensure_all_tables_registered() in the shared tests/unit/manager/repositories/conftest.py.
Repository tests register only part of the model graph transitively (e.g. DomainRow via query scopes). SQLAlchemy's global configure_mappers() then fails to resolve string relationships across the rest of the graph (DomainRow → "ScalingGroupForDomainRow" → "AgentRow" → …). On main this passed only by lucky Pants shard composition; adding a new test target in this PR reshuffled the shards and exposed it. Eager full-model registration makes mapper configuration independent of test sharding. (Verified: the affected repositories/base|rbac|ops tests fail on pure main in isolation too.)

Test plan

pants fmt / fix / lint / check / test on changed targets
Previously-failing repositories/base|rbac|ops tests + idle_checker tests pass locally
Full --changed-since suite passes via the pre-push hook
CI green

Resolves BA-6581

🤖 Generated with Claude Code

Copilot

Pull request overview

This PR replaces the placeholder idle-check reconciler source with a real data path. It introduces an IdleCheckerRepository (backed by a new IdleCheckerDBSource) that fetches a normalized IdleCheckSnapshot — sessions, their applicable scopes (resource group / project / domain), scope→checker bindings, and typed checker specs — and wires this repository through the orchestration DI and the reconciler stage factory so IdleCheckSource.fetch_reconcile_info returns the snapshot instead of an empty placeholder. It also makes IdleCheckerSpecABC.load dispatch by checker_type into concrete (currently field-less) spec classes that land their fields in follow-up stories. The change is read-only; the handler/applier remain placeholders.

Changes:

New idle_checker repository layer (repository, repositories, db_source, result types) producing IdleCheckSnapshot, plus new ScopeType/ScopeRef/IdleCheckSessionView data types.
DI wiring of idle_checker_repository through Repositories, both orchestration composers, sokovan input, and the reconciler stage factory, with IdleCheckReconcileInfo switched from session_ids to snapshot.
IdleCheckerSpecABC.load now dispatches on checker_type to SessionLifetimeSpec / NetworkTimeoutSpec / UtilizationSpec.

Reviewed changes

Copilot reviewed 19 out of 21 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
`repositories/idle_checker/db_source/db_source.py`	New DB reads assembling the normalized snapshot (joins + correlated enabled-binding filter).
`repositories/idle_checker/repository.py`	Thin repository delegating to the db_source.
`repositories/idle_checker/repositories.py`	`IdleCheckerRepositories` container following the per-entity wrapper convention.
`repositories/idle_checker/types.py`	`IdleCheckerRef`, `IdleCheckerBindingRef`, `IdleCheckSnapshot` result types.
`repositories/repositories.py`	Registers `idle_checker` in the aggregate `Repositories`.
`data/idle_checker/types.py`	Adds `ScopeType`, `ScopeRef`, `IdleCheckSessionView`.
`models/idle_checker/spec.py`	`load` discriminated dispatch + concrete spec subclasses.
`sokovan/idle_check/source.py`	Injects repository; fetches real snapshot.
`sokovan/idle_check/types.py`	`IdleCheckReconcileInfo` carries `snapshot`; `entity_ids` derives from it.
`sokovan/stages/idle_check.py`	`build_idle_check_stage` now takes the repository.
`sokovan/stages/factory.py`	Threads `idle_checker_repository` into stage assembly.
`dependencies/orchestration/composer.py`, `dependencies/orchestration/sokovan.py`, `dependencies/composer.py`	DI plumbing of the new repository.
`tests/.../idle_check/test_source.py`, `test_stage.py`, `dependencies/orchestration/test_sokovan.py`, `test_composer.py`	Update/add tests for the new repository argument and source behavior.
`changes/12398.feature.md`	Changelog entry.

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

+    async def fetch_idle_check_snapshot(
+        self, session_statuses: Collection[SessionStatus]
+    ) -> IdleCheckSnapshot:
+        """Fetch sessions and scope-bound idle checkers without expanding to session-checker rows."""


HyeockJinKim · 2026-06-29T10:42:19Z

+            sa.select(
+                SessionRow.id.label("session_id"),
+                SessionRow.created_at.label("session_created_at"),
+                SessionRow.starts_at.label("session_starts_at"),
+                ScalingGroupRow.id.label("resource_group_id"),
+                SessionRow.group_id.label("project_id"),
+                DomainRow.id.label("domain_id"),
+            )
+            .select_from(SessionRow)
+            .join(DomainRow, SessionRow.domain_name == DomainRow.name)
+            .join(ScalingGroupRow, SessionRow.scaling_group_name == ScalingGroupRow.name)


Is there a reason to use a JOIN like this?

This is to look up the ID in Scope. Since the current session row only contains the domain name and resource group name, a join is required to determine their IDs.

jopemachine · 2026-07-01T01:54:39Z

If possible, I don't want to comment on naming, but wasn't it the convention to use the Data suffix for these data pattern types?

Is there a reason for omitting the Data suffix here?

I think it would be clearer if the type's purpose could be identified from its name alone.

fregataa · 2026-07-03T07:26:32Z

+                    # today that is RUNNING only. Kept as a parameter so the policy stays in the
+                    # stage's target_statuses rather than baked into this read.
+                    SessionRow.status.in_(session_statuses),
+                    SessionRow.session_type != SessionTypes.INFERENCE,


This condition (session_type) also can be a reconcile target

…source Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…tainer Construct it in an IdleCheckerRepositories container and pass it through OrchestrationInput like the other repositories, instead of building it inline in the orchestration composer. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…k read - Read resource_group_id/domain_id from SessionRow columns (BA-6644) instead of joining DomainRow/ScalingGroupRow in the batch fetch, correlating the enabled-binding EXISTS subquery on those columns too - Move the session_type filter into the stage's target_statuses as session_types, matching how session_statuses is owned by the stage - Populate the new NOT NULL scope-id columns in the repository fixtures Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

HyeockJinKim · 2026-07-03T09:23:10Z

+            # Inference sessions follow the deployment/routing lifecycle, not idle termination.
+            session_types=frozenset({
+                SessionTypes.INTERACTIVE,
+                SessionTypes.BATCH,
+                SessionTypes.SYSTEM,
+            }),


Why is the system session targeted?

HyeockJinKim · 2026-07-03T09:26:17Z

+    def _enabled_binding_exists_query(self) -> sa.sql.elements.ColumnElement[bool]:
+        return sa.exists().where(
+            sa.and_(
+                IdleCheckerBindingRow.enabled == sa.true(),
+                sa.or_(
+                    sa.and_(
+                        IdleCheckerBindingRow.scope_type == ScopeType.RESOURCE_GROUP.value,
+                        IdleCheckerBindingRow.scope_id == SessionRow.resource_group_id,
+                    ),
+                    sa.and_(
+                        IdleCheckerBindingRow.scope_type == ScopeType.PROJECT.value,
+                        IdleCheckerBindingRow.scope_id == SessionRow.group_id,
+                    ),
+                    sa.and_(
+                        IdleCheckerBindingRow.scope_type == ScopeType.DOMAIN.value,
+                        IdleCheckerBindingRow.scope_id == SessionRow.domain_id,
+                    ),
+                ),
+            )
+        )


I'm a little disappointed with the query binding part—wouldn't it make more sense to query based on IdleChecker or the session?

github-actions Bot added the size:L 100~500 LoC label Jun 24, 2026

github-actions Bot assigned seedspirit Jun 24, 2026

github-actions Bot added the comp:manager Related to Manager component label Jun 24, 2026

seedspirit added a commit that referenced this pull request Jun 24, 2026

changelog: add news fragment for PR #12398

c159b94

seedspirit marked this pull request as ready for review June 24, 2026 09:37

seedspirit requested a review from a team as a code owner June 24, 2026 09:37

Copilot AI review requested due to automatic review settings June 24, 2026 09:37

Copilot started reviewing on behalf of seedspirit June 24, 2026 09:38 View session

Copilot AI reviewed Jun 24, 2026

View reviewed changes

seedspirit marked this pull request as draft June 24, 2026 10:14

seedspirit marked this pull request as ready for review June 25, 2026 01:54

seedspirit requested a review from HyeockJinKim June 25, 2026 02:02

github-actions Bot added size:XL 500~ LoC comp:common Related to Common component and removed size:L 100~500 LoC labels Jun 29, 2026

seedspirit added a commit that referenced this pull request Jun 29, 2026

changelog: add news fragment for PR #12398

2c94146

seedspirit force-pushed the feat/BA-6581 branch from 25ce1ff to afb0e70 Compare June 29, 2026 09:18

HyeockJinKim reviewed Jun 29, 2026

View reviewed changes

Comment thread src/ai/backend/common/data/idle_checker/types.py Outdated

HyeockJinKim reviewed Jun 29, 2026

View reviewed changes

seedspirit added a commit that referenced this pull request Jun 30, 2026

changelog: add news fragment for PR #12398

7282735

seedspirit force-pushed the feat/BA-6581 branch from aa735bd to be632fa Compare June 30, 2026 01:47

github-actions Bot added size:L 100~500 LoC size:XL 500~ LoC and removed size:XL 500~ LoC size:L 100~500 LoC labels Jun 30, 2026

jopemachine reviewed Jul 1, 2026

View reviewed changes

Comment thread src/ai/backend/manager/data/idle_checker/types.py

jopemachine reviewed Jul 1, 2026

View reviewed changes

jopemachine previously approved these changes Jul 1, 2026

View reviewed changes

seedspirit added a commit that referenced this pull request Jul 3, 2026

changelog: add news fragment for PR #12398

70ae8db

seedspirit dismissed jopemachine’s stale review via 1084e15 July 3, 2026 05:51

seedspirit force-pushed the feat/BA-6581 branch 5 times, most recently from 1869788 to 9a884c5 Compare July 3, 2026 06:44

fregataa previously approved these changes Jul 3, 2026

View reviewed changes

seedspirit and others added 9 commits July 3, 2026 17:32

feat(BA-6581): wire IdleCheckerRepository into idle-check reconciler …

d1f2c3e

…source Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

changelog: add news fragment for PR #12398

a0daf9b

refactor(BA-6581): store idle-checker spec via PydanticColumn

cee60bd

refactor(BA-6581): expose idle-check source as batch targets

3d1e71e

refactor: suffix idle-check repository data types

942a15d

refactor: use ops provider for idle-check reads

370071e

test: cover idle-check repository batch fetch

288f792

seedspirit dismissed fregataa’s stale review via db79915 July 3, 2026 09:09

seedspirit force-pushed the feat/BA-6581 branch from 9a884c5 to db79915 Compare July 3, 2026 09:09

HyeockJinKim reviewed Jul 3, 2026

View reviewed changes

fregataa approved these changes Jul 3, 2026

View reviewed changes

HyeockJinKim reviewed Jul 3, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

feat(BA-6581): wire IdleCheckerRepository into idle-check reconciler source#12398

feat(BA-6581): wire IdleCheckerRepository into idle-check reconciler source#12398
seedspirit wants to merge 9 commits into
mainfrom
feat/BA-6581

seedspirit commented Jun 24, 2026 •

edited

Loading

Uh oh!

Copilot AI left a comment

Uh oh!

Uh oh!

HyeockJinKim Jun 29, 2026

Uh oh!

seedspirit Jun 30, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jopemachine Jul 1, 2026

Uh oh!

fregataa Jul 3, 2026

Uh oh!

HyeockJinKim Jul 3, 2026

Uh oh!

HyeockJinKim Jul 3, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

Uh oh!

Conversation

seedspirit commented Jun 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Summary

Incidental test-infra fix

Test plan

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

HyeockJinKim Jun 29, 2026

Choose a reason for hiding this comment

Uh oh!

seedspirit Jun 30, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jopemachine Jul 1, 2026

Choose a reason for hiding this comment

Uh oh!

fregataa Jul 3, 2026

Choose a reason for hiding this comment

Uh oh!

HyeockJinKim Jul 3, 2026

Choose a reason for hiding this comment

Uh oh!

HyeockJinKim Jul 3, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

seedspirit commented Jun 24, 2026 •

edited

Loading