feat: serving classes v1 by hachall · Pull Request #1774 · doublewordai/control-layer

hachall · 2026-09-14T11:55:42Z

Preview build for serving classes v1 (dispatch-mode resolution in onwards, pool tag + priority envelope to the self-hosted serving stack, account-level customisation, catalog-declared active classes and overlays).

Everything is dormant by default: no model activates a class and no account is granted one, so every request resolves to standard and upstream bytes are unchanged until a model and an organisation are opted in. This branch exists to exercise default and customised behaviour in a preview environment against a database fork.

Design notes and rollout plan are tracked internally; the commit message carries the component-level summary.

🤖 Generated with Claude Code

…the pool envelope Adds serving classes v1: a per-request dispatch mode (`interactive`, `throughput`, or the default `standard`) resolved in onwards and carried to the self-hosted serving stack as a pool tag plus priority band. Everything ships dormant: no model activates a class and no account is granted one, so every request resolves to `standard` and the bytes reaching upstreams are unchanged until an operator opts a model and an organisation in. Onwards - `serving` module: class type, suffix parsing (`alias:class`, unknown class is a 400), key policy (`KeyServing`: key class, account default, no external failover, per-model overlays), and the pure resolver (suffix > key > overlay default > account default; gated by the alias's active classes and the org's grant; batch and continuation purposes always resolve to standard). - Handler: resolves after auth and routing rules, records requested and resolved class on the span, a counter and a `ServingClassOutcome` response extension; stamps `x-dynamo-interactivity-pool` and `x-dynamo-request-priority` only on members that accept scheduling fields and only for elevated classes; skips untrusted (external) members for self-hosted-only accounts, consuming the attempt budget so the request never falls over externally. - Config: `PoolSpec.serving_classes` on the default pool, `KeyDefinition.serving`, `Targets.key_serving` kept in sync on reload. dwctl - Migration 141: `deployed_models.serving_classes`, `api_keys.serving_class` (in the NOTIFY column set), `users.default_serving_class` and `users.self_hosted_only` (account settings next to ZDR, with a NOTIFY trigger), `model_overlays` (org × model: tariff name, granted classes, default class, routing override, provisioning source), and the two analytics columns. - Inference middleware strips the class suffix at the outermost layer so analytics, the prompt cache and billing key on the bare alias, passes the class to onwards as a request extension, and scrubs the client-side envelope (headers and `nvext.interactivity_pool`) alongside the existing priority strip. - Onwards sync emits the per-key serving policy (omitted when empty) and the per-alias active classes. - Analytics rows record requested and resolved class. - API: `serving_class` on API keys; `default_serving_class` and `self_hosted_only` on users and organisations (UpdateAll or org owner); `serving_classes` on model responses. - Model catalog: `clay.serving_classes` and `clay.overlays` (org by account username, tariff, classes, default class, routing override), validated and materialised into `model_overlays`; JSON schema regenerated. Not yet: pricing by overlay tariff, legacy alias rewrite, console views, fusillade `retry_after_ms`, dashboards.

cloudflare-workers-and-pages · 2026-09-14T11:56:17Z

Deploying control-layer with Cloudflare Pages

Latest commit:	`c96393f`
Status:	✅ Deploy successful!
Preview URL:	https://5b326000.control-layer.pages.dev
Branch Preview URL:	https://preview-serving-classes.control-layer.pages.dev

View logs

# Conflicts: # dwctl/src/db/handlers/api_keys.rs

# Conflicts: # dwctl/src/request_logging/batcher.rs

…verlay files and endpoint kind Reshapes v1 after review: - Per-key serving class removed. Precedence is request suffix, then the organisation's per-model overlay default, then its account default. - An elevated class applies when the organisation holds it (account setting `granted_serving_classes`) and the model offers it (`serving_classes` in the model catalog). Strict mode: a class named on the request that is not held or not offered is refused with 403 `serving_class_unavailable`; only the silent defaults degrade to `standard`. - Account settings (granted classes, default class, self-hosted-only) live on users and organisations and are set through the API, gated to UpdateAll or the organisation owner. - Per-organisation overlay catalog (`org_overlays`): one YAML per organisation carrying per-model overrides only (default class, routing), applied after the model catalog; organisations only; rows are owned by the file and pruned when no longer declared. - Explicit endpoint kind (`dynamo`, `hosted`, `external`) on inference endpoints, backfilled from the scheduling capability, exposed on the endpoint API. The envelope is stamped only on dynamo members and the self-hosted-only restriction skips only external members; neither reads another flag any more. - Onwards config carries accounts once per organisation and overlays on the alias's pool spec; every key gains an `account` label. Unaffected keys' definitions are unchanged. - Migration renumbered to 143 after main's 142.

# Conflicts: # dwctl/src/db/handlers/deployments.rs # dwctl/src/db/handlers/model_provisioning.rs # dwctl/src/sync/onwards_config/tests.rs # onwards/src/handlers.rs

Inference lab is moving from forked interactivity pools to dynamo's GlobalRouter and GlobalPlanner, which place a request by objective targets rather than a pool name. Serving classes follow: - A model's `serving_classes` is now a map of class -> preset `{ttft_ms, itl_ms, priority}` (JSONB; catalog `clay.serving_classes:`). Offering a class = declaring its preset; an optional `standard` preset pins what unclassed traffic asks for, otherwise standard sends nothing. - Onwards writes the resolved preset into the body of `dynamo`-kind members as `nvext.router.{ttft_target, itl_target}`, and the priority as `nvext.agent_hints.priority` only when non-zero. The two envelope headers and the fixed priority bands are gone. A resolution without targets leaves the body untouched. - An org overlay names either `default_class` or explicit `targets` (CHECK and validator refuse both). Explicit targets apply as-is when the request names no class, imply authority (no grant check) and resolve as `custom`. - dwctl's ingress scrubs client-sent `nvext.router` targets, alongside the body priority, and no longer strips the removed headers. - Migration renumbered to 147 (main took 146).

Copilot

🟡 Changes recommended

Request scrubbing can be bypassed, organization owners can self-grant elevated classes, and generated contracts and SQLx metadata are inconsistent.

Get a fresh assessment by requesting another Copilot review.

Pull request overview

Introduces serving-class routing, account/model overlays, endpoint classification, and analytics persistence across onwards and dwctl.

Changes:

Adds serving-class resolution and provider routing metadata.
Adds database/API support for account grants, overlays, presets, and endpoint kinds.
Persists serving-class analytics and refreshes SQLx metadata.

File summaries

File	Description
`onwards/src/strict/mod.rs`	Updates test target state.
`onwards/src/load_balancer.rs`	Stores serving presets and overlays.
`model-provisioning.schema.json`	Defines serving-preset schema.
`dwctl/src/test/utils.rs`	Updates user fixtures.
`dwctl/src/sync/endpoint_sync.rs`	Updates sync fixtures.
`dwctl/src/sample_files/mod.rs`	Updates deployment fixture.
`dwctl/src/request_logging/serializers.rs`	Extracts serving outcomes.
`dwctl/src/request_logging/batcher.rs`	Persists serving analytics.
`dwctl/src/request_logging/analytics_handler.rs`	Propagates serving metrics.
`dwctl/src/metrics/cache_info.rs`	Updates endpoint fixtures.
`dwctl/src/lib.rs`	Applies organization overlays.
`dwctl/src/db/models/users.rs`	Adds account serving fields.
`dwctl/src/db/models/organizations.rs`	Adds organization serving updates.
`dwctl/src/db/models/inference_endpoints.rs`	Adds endpoint kinds.
`dwctl/src/db/models/deployments.rs`	Adds serving presets.
`dwctl/src/db/handlers/organizations.rs`	Persists organization settings.
`dwctl/src/db/handlers/model_provisioning.rs`	Provisions model presets.
`dwctl/src/db/handlers/deployments.rs`	Projects serving presets.
`dwctl/src/config.rs`	Configures overlay directory.
`dwctl/src/auth/middleware.rs`	Updates endpoint fixtures.
`dwctl/src/auth/current_user.rs`	Updates user-update fixture.
`dwctl/src/api/models/users.rs`	Exposes user serving settings.
`dwctl/src/api/models/organizations.rs`	Exposes organization settings.
`dwctl/src/api/models/inference_endpoints.rs`	Exposes endpoint kinds.
`dwctl/src/api/models/deployments/mod.rs`	Exposes model presets.
`dwctl/src/api/models/deployments/enrichment.rs`	Updates enrichment fixtures.
`dwctl/src/api/handlers/users.rs`	Validates serving updates.
`dwctl/src/api/handlers/payments.rs`	Updates account mutations.
`dwctl/src/api/handlers/organizations.rs`	Authorizes organization settings.
`dwctl/src/api/handlers/mod.rs`	Adds class validation.
`dwctl/src/api/handlers/inference_endpoints.rs`	Persists endpoint kinds.
`dwctl/src/api/handlers/auth.rs`	Updates user mutations.
`dwctl/migrations/147_serving_classes.sql`	Creates serving-class schema.
`.sqlx/query-f6313c….json`	API-key insert metadata.
`.sqlx/query-e21376….json`	API-key update metadata.
`.sqlx/query-d9bbd3….json`	Stale analytics metadata.
`.sqlx/query-d57c49….json`	Endpoint bulk-query metadata.
`.sqlx/query-cc71ec….json`	Overlay-query metadata.
`.sqlx/query-c28831….json`	Organization-query metadata.
`.sqlx/query-c1deed….json`	User-query metadata.
`.sqlx/query-c1dbc0….json`	Organization-insert metadata.
`.sqlx/query-b26923….json`	Organization-insert metadata.
`.sqlx/query-a8267e….json`	Account-policy metadata.
`.sqlx/query-a3d4a7….json`	Endpoint-list metadata.
`.sqlx/query-a341d2….json`	User-query metadata.
`.sqlx/query-a289a4….json`	User-update metadata.
`.sqlx/query-9fbf45….json`	Organization-update metadata.
`.sqlx/query-8ad0f8….json`	Removes superseded metadata.
`.sqlx/query-880598….json`	API-key bulk metadata.
`.sqlx/query-847c27….json`	Organization-update metadata.
`.sqlx/query-7e5b73….json`	Composite-key metadata.
`.sqlx/query-718d03….json`	Endpoint-insert metadata.
`.sqlx/query-71802b….json`	Endpoint-get metadata.
`.sqlx/query-6cedb5….json`	User-query metadata.
`.sqlx/query-6c53c3….json`	Composite-key metadata.
`.sqlx/query-5e2e7b….json`	Overlay-query metadata.
`.sqlx/query-58ce43….json`	Composite-model metadata.
`.sqlx/query-5849bc….json`	Organization-query metadata.
`.sqlx/query-42d523….json`	Overlay-query metadata.
`.sqlx/query-3eeeff….json`	User-query metadata.
`.sqlx/query-39b331….json`	Analytics-insert metadata.
`.sqlx/query-3899b1….json`	API-key get metadata.
`.sqlx/query-2f988a….json`	API-key update metadata.
`.sqlx/query-2dac89….json`	API-key insert metadata.
`.sqlx/query-185aa2….json`	Composite-model metadata.
`.sqlx/query-17af29….json`	API-key list metadata.
`.sqlx/query-0eac1f….json`	User-insert metadata.
`.sqlx/query-0e8ba1….json`	Endpoint-update metadata.

Review details

Files not reviewed (43)

.sqlx/query-005684c6b3cb520548bd107621b475cff7981f3929f0d9e73252f339765cd2dc.json: Generated file
.sqlx/query-0d04b215ce9a85392b8805e2fd8ab7a6cde6b040bbe36461d9404a937b9f2593.json: Generated file
.sqlx/query-0e8ba17bbd07d3aba6fa3c89a357bdc0b1544bbaa853d0ddcaff60792fec0b65.json: Generated file
.sqlx/query-17af2947bdf088fa1d831c7408139f43bed741349b93a5d6757e88ac1945efc2.json: Generated file
.sqlx/query-185aa2bcf4edcfcdd404d62b4052d1c76dc72c04f15cd97b1be6eb5e47fa9baa.json: Generated file
.sqlx/query-1e58059d2038426045c82e58c7eb452fae8f00be38bd884b57ceeef70dbcc8cf.json: Generated file
.sqlx/query-2c443be50801c391b4323ed83847fdc4e4fb747cb5465a74bdfd451b1549aa14.json: Generated file
.sqlx/query-2dac896d9529ee0ee4bdf09851a565307a6a3354da26e7f493447072da8cdffc.json: Generated file
.sqlx/query-2f988ac0d6052a573b0277fdc137f3a16f73e5292adf7b913272c0e3a1774dca.json: Generated file
.sqlx/query-3760af261851960f785fa7318d9567dfe90150961bad41ce15f9e0c9423eefdc.json: Generated file
.sqlx/query-3899b1a854d897ffb6ab1a7f44e351278f9623c4f46247a034e0a592faafafa8.json: Generated file
.sqlx/query-39b331b3b9ec624cf01444e3c065bc5dd2fdf6d9f998dbe709c6ae0f49f80b3f.json: Generated file
.sqlx/query-3eeeff204fd4edb60c30709611619a9005bdb5fd4f1fdd8e87faa4322f295677.json: Generated file
.sqlx/query-42d523ce78860a8358b81d5f8d43fcacce9f59c7c4259f60331b31f86069e473.json: Generated file
.sqlx/query-5849bc1e94bb696c829b8b37f792a908da49c92f3e2b8e3dbaa3a601473d13b2.json: Generated file
.sqlx/query-5bee3bad4a63e4b46a016115a1776622affe52125e2c1364284f1d56a27c6b4d.json: Generated file
.sqlx/query-5e2e7b9e79dbaa73ed44d626b8128ba65082fbae6fcc141f641c74e585f0d5dd.json: Generated file
.sqlx/query-66494123fee88db52c985676a783eeac92cc99165ed65804b45802a6b697ce1a.json: Generated file
.sqlx/query-66d87d1c182fc66f447a8da5b796ba82f106d53df4da8b936f05f90e70d4ab9b.json: Generated file
.sqlx/query-6c53c32ae51e6c9f8fe76429af124f964e24cb6896aafa928065ae77e5f6eaf9.json: Generated file
.sqlx/query-6dc1f2e8e5ee7eb2cd49a3890a46537bf6e0be92906fabe93efa6e59efda8266.json: Generated file
.sqlx/query-71802b819175b2e644063162a4bcba1c5f49c61148211716725bbcb3175d46a9.json: Generated file
.sqlx/query-718d03f3e12dafc9d9bd34f144bfb5279bf64ea292875d33012e85af1c8e848c.json: Generated file
.sqlx/query-75b7faa96dc55e600008464fa10d84280884266cc88e4a2b6a7f9df4595e03b3.json: Generated file
.sqlx/query-7e5b73f7064b6ddf18c203e68623be7543d7376b69f0922f2390da7aa4cb6a2b.json: Generated file
.sqlx/query-847c277e34b28ff500376f7140e61c88ee23e596824a257a91565b3da079cae3.json: Generated file
.sqlx/query-880598e51f20cae01484f689084dff474ff1bc5a65233fd21050da8fe4b279b6.json: Generated file
.sqlx/query-8ad0f85894651acb48f8c2c48eaf9277afd20b5ea8a8aff5b616e33bbeb21e1c.json: Generated file
.sqlx/query-9fbf45082890e8ce04783e093ed014d5e4df6a2b52278a807fd90419e941fb64.json: Generated file
.sqlx/query-a341d2e54b667a7020e013cb6fdbb6bf6ec887e855399f6175330776cfe01a40.json: Generated file
.sqlx/query-a3d4a784ffa4285fe7b54aa8d6f19e1a3b6865f9f2721684ffc53b6142164d4e.json: Generated file
.sqlx/query-a8267e1c6bf5ae1e38c2ddb155ba103831eb7aaa02edbdbe1bdd805600ee31c3.json: Generated file
.sqlx/query-b26923e64d58f98e50cac53a7ab94d0b6fc105e312b9ead96336a0f20b098c8a.json: Generated file
.sqlx/query-c1dbc09d5eeb32a77dada7b99b570ff30fcda3e277a785f2e14c8b4ca963bd8c.json: Generated file
.sqlx/query-c288315499617e5e980f2607631512f72e5ffe9c7096aa8ea92f43d20bf1f850.json: Generated file
.sqlx/query-cc71ec28d68425eca8840fe6f500f5dfe48d391711ad7095ae2344eee84d7253.json: Generated file
.sqlx/query-cde54777d647f1bda5e589d6dd6f94d074142297b4c7421765801f868b47fdb2.json: Generated file
.sqlx/query-d57c4928455820ad54035cadd826191eb71ac3985e7cae9ff67f616e6d6f135b.json: Generated file
.sqlx/query-d9bbd3aa686c2021e550f852a9850f19d3154475192d177c8de61bc8d6d615f6.json: Generated file
.sqlx/query-e08ffcd23ac10fd1d7d195a30da54a584b1954c88b5a64dcd93613b882e4e5eb.json: Generated file
.sqlx/query-e21376d172135d01534230b034c94a6ff3bd860cce49536343b5622505b82930.json: Generated file
.sqlx/query-e734f12eb9058c6c11a6ece8a525b47114701ffcdc87afc39f99c2c140ae71c6.json: Generated file
.sqlx/query-f6313c114e1306c743699ff40320de7e61fcecdb66e8f0efd5b85c22df799468.json: Generated file

Files reviewed: 50/93 changed files
Comments generated: 5
Review effort level: Balanced

💡 Configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

cubic-dev-ai

Review completed against the latest diff

_{Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic}

… catalog schema - Inference middleware: every scrubber now reports whether it removed something, and the realtime body is rebuilt whenever any of them did. Onwards forwards the bytes it receives, so a top-level or nvext scheduling priority (and a client-supplied response id) removed only from the parsed value would otherwise have reached the upstream unchanged. - Organisations: granted serving classes require UpdateAll (platform operator), matching the users endpoint; the default class and routing preference stay owner-settable like ZDR. Test added. - Catalog schema: ttft_ms and itl_ms carry minimum 1, matching the runtime validator. - Analytics doc comment: precedence no longer mentions a per-key class.

cubic-dev-ai

All reported issues were addressed across 68 files

_{Requires human review: Auto-approval blocked because this review re-detected 9 unresolved issues already reported by Cubic.
Tip: instead of fixing issues one by one fix them all with cubic

Re-trigger cubic}

onwards: - Resolve presets and overlays from the alias's default pool whichever named pool serves the request; a completions-pool request no longer sees an empty policy. - A self-hosted-only account's external members are excluded before the attempt budget is spent, under every strategy; an alias with no eligible member is refused as 503 `no_eligible_provider` rather than 429. - The serving outcome extension also rides on upstream-error responses so analytics records the resolved class for failed requests. - A class suffix on the selected model string (Model-Override header) outranks the body-derived extension. - `custom` is an outcome only: not deserialisable from configuration. - Alias serving policy is shared (Arc) so per-request pool clones do not scale with the number of organisations configured. dwctl: - Overlays of soft-deleted accounts are not synced. - Org-overlay apply takes a transaction advisory lock; both catalogs are parsed and validated before either is applied. - An overlay path that exists but is not a directory fails startup. - Outbox rows written before the serving-class fields existed still deserialise. - Endpoint create derives `kind` from `accepts_scheduling_priority` when omitted; endpoint kind round-trip test added. - Migration: local lock_timeout; column comment lists `custom`. - Config reference documents `org_overlays_directory`.

cubic-dev-ai

All reported issues were addressed across 21 files (changes from recent commits).

_{Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic}

Copilot

🟡 Changes recommended

Model-override precedence, incomplete analytics outcomes, and non-atomic catalog application can produce incorrect or transient routing behavior.

Get a fresh assessment by requesting another Copilot review.

Review details

Files not reviewed (18)

.sqlx/query-0d04b215ce9a85392b8805e2fd8ab7a6cde6b040bbe36461d9404a937b9f2593.json: Generated file
.sqlx/query-0e8ba17bbd07d3aba6fa3c89a357bdc0b1544bbaa853d0ddcaff60792fec0b65.json: Generated file
.sqlx/query-2812e4f2b125ba6710354865fa125d23761fc8f43b2a37f3e111369eae847349.json: Generated file
.sqlx/query-3760af261851960f785fa7318d9567dfe90150961bad41ce15f9e0c9423eefdc.json: Generated file
.sqlx/query-39b331b3b9ec624cf01444e3c065bc5dd2fdf6d9f998dbe709c6ae0f49f80b3f.json: Generated file
.sqlx/query-5849bc1e94bb696c829b8b37f792a908da49c92f3e2b8e3dbaa3a601473d13b2.json: Generated file
.sqlx/query-5bee3bad4a63e4b46a016115a1776622affe52125e2c1364284f1d56a27c6b4d.json: Generated file
.sqlx/query-66d87d1c182fc66f447a8da5b796ba82f106d53df4da8b936f05f90e70d4ab9b.json: Generated file
.sqlx/query-6c53c32ae51e6c9f8fe76429af124f964e24cb6896aafa928065ae77e5f6eaf9.json: Generated file
.sqlx/query-71802b819175b2e644063162a4bcba1c5f49c61148211716725bbcb3175d46a9.json: Generated file
.sqlx/query-718d03f3e12dafc9d9bd34f144bfb5279bf64ea292875d33012e85af1c8e848c.json: Generated file
.sqlx/query-75b7faa96dc55e600008464fa10d84280884266cc88e4a2b6a7f9df4595e03b3.json: Generated file
.sqlx/query-8ad0f85894651acb48f8c2c48eaf9277afd20b5ea8a8aff5b616e33bbeb21e1c.json: Generated file
.sqlx/query-9fbf45082890e8ce04783e093ed014d5e4df6a2b52278a807fd90419e941fb64.json: Generated file
.sqlx/query-a3d4a784ffa4285fe7b54aa8d6f19e1a3b6865f9f2721684ffc53b6142164d4e.json: Generated file
.sqlx/query-a8267e1c6bf5ae1e38c2ddb155ba103831eb7aaa02edbdbe1bdd805600ee31c3.json: Generated file
.sqlx/query-b26923e64d58f98e50cac53a7ab94d0b6fc105e312b9ead96336a0f20b098c8a.json: Generated file
.sqlx/query-d57c4928455820ad54035cadd826191eb71ac3985e7cae9ff67f616e6d6f135b.json: Generated file

Files reviewed: 51/69 changed files
Comments generated: 4
Review effort level: Balanced

…rror type, clippy - Model responses include `serving_classes` only for callers with Models::ReadAll: the presets are an operator detail, never shown to customers. - The serving outcome travels on OnwardsErrorResponse and is attached as a response extension on conversion, so error handling (including strict mode's) is unchanged and analytics still sees the class on failures. - clippy (newer toolchain in CI): redundant closure in the overlay upsert.

Copilot

🟡 Changes recommended

Model-override precedence and post-resolution error paths can still produce incorrect serving-class attribution, and overlay ownership documentation is inconsistent.

Get a fresh assessment by requesting another Copilot review.

Review details

Files not reviewed (18)

.sqlx/query-0d04b215ce9a85392b8805e2fd8ab7a6cde6b040bbe36461d9404a937b9f2593.json: Generated file
.sqlx/query-0e8ba17bbd07d3aba6fa3c89a357bdc0b1544bbaa853d0ddcaff60792fec0b65.json: Generated file
.sqlx/query-2812e4f2b125ba6710354865fa125d23761fc8f43b2a37f3e111369eae847349.json: Generated file
.sqlx/query-3760af261851960f785fa7318d9567dfe90150961bad41ce15f9e0c9423eefdc.json: Generated file
.sqlx/query-39b331b3b9ec624cf01444e3c065bc5dd2fdf6d9f998dbe709c6ae0f49f80b3f.json: Generated file
.sqlx/query-5849bc1e94bb696c829b8b37f792a908da49c92f3e2b8e3dbaa3a601473d13b2.json: Generated file
.sqlx/query-5bee3bad4a63e4b46a016115a1776622affe52125e2c1364284f1d56a27c6b4d.json: Generated file
.sqlx/query-66d87d1c182fc66f447a8da5b796ba82f106d53df4da8b936f05f90e70d4ab9b.json: Generated file
.sqlx/query-6c53c32ae51e6c9f8fe76429af124f964e24cb6896aafa928065ae77e5f6eaf9.json: Generated file
.sqlx/query-71802b819175b2e644063162a4bcba1c5f49c61148211716725bbcb3175d46a9.json: Generated file
.sqlx/query-718d03f3e12dafc9d9bd34f144bfb5279bf64ea292875d33012e85af1c8e848c.json: Generated file
.sqlx/query-75b7faa96dc55e600008464fa10d84280884266cc88e4a2b6a7f9df4595e03b3.json: Generated file
.sqlx/query-8ad0f85894651acb48f8c2c48eaf9277afd20b5ea8a8aff5b616e33bbeb21e1c.json: Generated file
.sqlx/query-9fbf45082890e8ce04783e093ed014d5e4df6a2b52278a807fd90419e941fb64.json: Generated file
.sqlx/query-a3d4a784ffa4285fe7b54aa8d6f19e1a3b6865f9f2721684ffc53b6142164d4e.json: Generated file
.sqlx/query-a8267e1c6bf5ae1e38c2ddb155ba103831eb7aaa02edbdbe1bdd805600ee31c3.json: Generated file
.sqlx/query-b26923e64d58f98e50cac53a7ab94d0b6fc105e312b9ead96336a0f20b098c8a.json: Generated file
.sqlx/query-d57c4928455820ad54035cadd826191eb71ac3985e7cae9ff67f616e6d6f135b.json: Generated file

Files reviewed: 53/71 changed files
Comments generated: 1
Review effort level: Balanced

+//! Rows written here are owned by the file (`provisioning_source`) and are
+//! rewritten on every start, exactly like the model catalog; a row the
+//! catalog does not mention is left alone.


cubic-dev-ai

1 issue found across 7 files (changes from recent commits).

Prompt for AI agents (unresolved issues)


Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="dwctl/src/api/models/deployments/mod.rs">

<violation number="1" location="dwctl/src/api/models/deployments/mod.rs:701">
P2: Platform-manager create and update responses now drop `serving_classes` because they use this converter without the privileged enrichment applied by list/get. Attach the presets in those responses too, so all `Models::ReadAll` model endpoints expose the same catalog metadata.</violation>
</file>

_{Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic}

cubic-dev-ai · 2026-09-18T09:18:01Z

                .filter(|m| *m != ModelCatalogMetadata::default()),
            provisioning_source: db.provisioning_source,
+            // Attached by the handler for platform managers only.
+            serving_classes: None,


P2: Platform-manager create and update responses now drop serving_classes because they use this converter without the privileged enrichment applied by list/get. Attach the presets in those responses too, so all Models::ReadAll model endpoints expose the same catalog metadata.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At dwctl/src/api/models/deployments/mod.rs, line 701: <comment>Platform-manager create and update responses now drop `serving_classes` because they use this converter without the privileged enrichment applied by list/get. Attach the presets in those responses too, so all `Models::ReadAll` model endpoints expose the same catalog metadata.</comment> <file context> @@ -700,7 +697,8 @@ impl From<DeploymentDBResponse> for DeployedModelResponse { provisioning_source: db.provisioning_source, - serving_classes: db.serving_classes, + // Attached by the handler for platform managers only. + serving_classes: None, } } </file context>

hachall added 2 commits September 14, 2026 13:03

Merge branch 'main' into preview/serving-classes

cb201fb

Merge branch 'main' into preview/serving-classes

e50e57e

# Conflicts: # dwctl/src/db/handlers/api_keys.rs

hachall changed the title ~~preview: serving classes v1~~ feat: serving classes v1 Sep 15, 2026

hachall added 8 commits September 15, 2026 11:09

Merge branch 'main' into preview/serving-classes

03b2038

Merge branch 'main' into preview/serving-classes

4b9fa06

Merge branch 'main' into preview/serving-classes

3762330

# Conflicts: # dwctl/src/request_logging/batcher.rs

Merge branch 'main' into preview/serving-classes

c12fb9b

# Conflicts: # dwctl/src/db/handlers/deployments.rs # dwctl/src/db/handlers/model_provisioning.rs # dwctl/src/sync/onwards_config/tests.rs # onwards/src/handlers.rs

chore: drop stray pycache file from the index

2a9da65

Merge branch 'main' into preview/serving-classes

ab6833e

hachall requested a balanced review from Copilot September 17, 2026 22:51

Merge branch 'main' into preview/serving-classes

ab63375

hachall marked this pull request as ready for review September 17, 2026 22:51

Copilot started reviewing on behalf of hachall September 17, 2026 22:51 View session

Copilot AI reviewed Sep 17, 2026

View reviewed changes

chore(sqlx): drop orphaned offline query data from earlier iterations

ba6c1cc

cubic-dev-ai Bot reviewed Sep 17, 2026

View reviewed changes

hachall requested a balanced review from Copilot September 17, 2026 23:55

Copilot started reviewing on behalf of hachall September 17, 2026 23:56 View session

cubic-dev-ai Bot reviewed Sep 18, 2026

View reviewed changes

Comment thread docs/src/reference/configuration.md

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Comment thread dwctl/src/inference/middleware.rs

Comment thread dwctl/src/lib.rs

Comment thread dwctl/src/request_logging/serializers.rs

Comment thread dwctl/src/api/models/organizations.rs Outdated

hachall added 3 commits September 18, 2026 09:57

docs(serving): organisation API comments say platform managers only

dd5c2a8

Merge branch 'main' into preview/serving-classes

c96393f

hachall requested a balanced review from Copilot September 18, 2026 09:08

Copilot started reviewing on behalf of hachall September 18, 2026 09:09 View session

Copilot AI reviewed Sep 18, 2026

View reviewed changes

Comment thread dwctl/src/org_overlays.rs

Comment on lines +12 to +14

//! Rows written here are owned by the file (`provisioning_source`) and are

//! rewritten on every start, exactly like the model catalog; a row the

//! catalog does not mention is left alone.

cubic-dev-ai Bot reviewed Sep 18, 2026

View reviewed changes

Conversation

hachall commented Sep 14, 2026

Uh oh!

cloudflare-workers-and-pages Bot commented Sep 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Deploying control-layer with Cloudflare Pages

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Files not reviewed (43)

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cubic-dev-ai Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Files not reviewed (18)

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

🟡 Changes recommended

Files not reviewed (18)

Uh oh!

cubic-dev-ai Bot left a comment

Choose a reason for hiding this comment

Uh oh!

cubic-dev-ai Bot Sep 18, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

cloudflare-workers-and-pages Bot commented Sep 14, 2026 •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot left a comment •

edited

Loading

cubic-dev-ai Bot Sep 18, 2026 •

edited

Loading