feat: generate RAG service YAML config resource#313
Conversation
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
📝 WalkthroughWalkthroughThis change introduces a new Changes
Poem
🚥 Pre-merge checks | ✅ 2 | ❌ 1❌ Failed checks (1 warning)
✅ Passed checks (2 passed)
✏️ Tip: You can configure your own custom pre-merge checks in the settings. ✨ Finishing Touches🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
@coderabbitai review |
|
✅ Actions performedReview triggered.
|
![coderabbitai[bot]](https://avatars.githubusercontent.com/in/347564?s=60&v=4){kind=small}
There was a problem hiding this comment.
Actionable comments posted: 2
Caution
Some comments are outside the diff and can’t be posted inline due to platform limitations.
⚠️ Outside diff range comments (1)
server/internal/orchestrator/swarm/orchestrator.go (1)
404-409:⚠️ Potential issue | 🟠 Major
raginstances are now provisionable but never deployed.
GenerateServiceInstanceResourcesroutes"rag"intogenerateRAGInstanceResources, but that path only returns filesystem/user/key/config resources. There is noServiceInstanceSpecResourceorServiceInstanceResource, so nothing ever creates a Swarm service for the persistedServiceInstance. In practice this leaves RAG instances stuck increatingwith no container that can ever become ready. Either keep the"rag"path disabled until deployment lands, or add the deployment resources in this PR.Also applies to: 531-594
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed. In `@server/internal/orchestrator/swarm/orchestrator.go` around lines 404 - 409, The "rag" branch in GenerateServiceInstanceResources currently calls generateRAGInstanceResources which only returns filesystem/user/key/config resources and omits ServiceInstanceSpecResource and ServiceInstanceResource, so RAG instances are never deployed; either remove/guard the "rag" case until deployment support lands, or modify generateRAGInstanceResources to create and return the same deployment resources as other service types (add ServiceInstanceSpecResource and ServiceInstanceResource entries, including container/service spec needed for Swarm creation) so the persisted ServiceInstance can be turned into a running Swarm service; update the switch in GenerateServiceInstanceResources and the body of generateRAGInstanceResources (and mirrored logic around lines 531–594) accordingly.
🤖 Prompt for all review comments with AI agents
Verify each finding against the current code and only fix it if needed.
Inline comments:
In `@server/internal/orchestrator/swarm/rag_config.go`:
- Around line 94-98: GenerateRAGConfig currently builds pipelines by calling
buildRAGPipelineYAML without validating that per-stage credentials match; this
lets a pipeline with EmbeddingLLM.APIKey and RAGLLM.APIKey set to different
values silently prefer one key and misauthenticate. Add a guard where pipelines
are iterated (in GenerateRAGConfig, and similarly in ParseRAGServiceConfig if
present) to detect when both EmbeddingLLM.APIKey and RAGLLM.APIKey are non-empty
for the same provider (e.g., "openai", "anthropic") and differ, then return an
error (fail fast) describing the mismatched keys and include the pipeline
identifier; update buildRAGPipelineYAML callers to rely on this validation so no
silent overwrite occurs.
- Around line 24-35: The YAML model is collapsing explicit zeros because
TokenBudget and TopN are plain ints with `omitempty`; update ragPipelineYAML to
use `*int` for TokenBudget and TopN (and change the corresponding numeric fields
in any related YAML structs such as ragDatabaseYAML and the RAGDefaults YAML
struct to `*int` as well), and preserve/propagate the pointers when mapping
between internal structs and these YAML structs so nil vs. explicit 0 is
preserved (look for all mappings that build/return ragPipelineYAML,
ragDatabaseYAML, and the RAGDefaults YAML structs and assign pointer values
instead of converting to plain ints).
---
Outside diff comments:
In `@server/internal/orchestrator/swarm/orchestrator.go`:
- Around line 404-409: The "rag" branch in GenerateServiceInstanceResources
currently calls generateRAGInstanceResources which only returns
filesystem/user/key/config resources and omits ServiceInstanceSpecResource and
ServiceInstanceResource, so RAG instances are never deployed; either
remove/guard the "rag" case until deployment support lands, or modify
generateRAGInstanceResources to create and return the same deployment resources
as other service types (add ServiceInstanceSpecResource and
ServiceInstanceResource entries, including container/service spec needed for
Swarm creation) so the persisted ServiceInstance can be turned into a running
Swarm service; update the switch in GenerateServiceInstanceResources and the
body of generateRAGInstanceResources (and mirrored logic around lines 531–594)
accordingly.
ℹ️ Review info
⚙️ Run configuration
Configuration used: Organization UI
Review profile: CHILL
Plan: Pro
Run ID: 2c6dcd4a-44fe-4d12-9fe1-c4b54b1b6cc8
📒 Files selected for processing (8)
server/internal/orchestrator/swarm/orchestrator.goserver/internal/orchestrator/swarm/rag_config.goserver/internal/orchestrator/swarm/rag_config_resource.goserver/internal/orchestrator/swarm/rag_config_resource_test.goserver/internal/orchestrator/swarm/rag_config_test.goserver/internal/orchestrator/swarm/rag_service_user_role_test.goserver/internal/orchestrator/swarm/resources.goserver/internal/orchestrator/swarm/service_spec.go
| type ragPipelineYAML struct { | ||
| Name string `yaml:"name"` | ||
| Description string `yaml:"description,omitempty"` | ||
| Database ragDatabaseYAML `yaml:"database"` | ||
| Tables []ragTableYAML `yaml:"tables"` | ||
| EmbeddingLLM ragLLMYAML `yaml:"embedding_llm"` | ||
| RAGLLM ragLLMYAML `yaml:"rag_llm"` | ||
| APIKeys *ragAPIKeysYAML `yaml:"api_keys,omitempty"` | ||
| TokenBudget int `yaml:"token_budget,omitempty"` | ||
| TopN int `yaml:"top_n,omitempty"` | ||
| SystemPrompt string `yaml:"system_prompt,omitempty"` | ||
| Search *ragSearchYAML `yaml:"search,omitempty"` |
There was a problem hiding this comment.
Preserve explicit zero values in the YAML model.
database.RAGPipeline.TokenBudget, TopN, and database.RAGDefaults are pointer fields, but the YAML structs flatten them to int with omitempty. That collapses an explicit 0 into “unset”, so the generated YAML no longer faithfully represents the parsed config. Keep the optional numeric fields as *int and pass the pointers through.
🔧 Suggested fix
type ragPipelineYAML struct {
Name string `yaml:"name"`
Description string `yaml:"description,omitempty"`
Database ragDatabaseYAML `yaml:"database"`
Tables []ragTableYAML `yaml:"tables"`
EmbeddingLLM ragLLMYAML `yaml:"embedding_llm"`
RAGLLM ragLLMYAML `yaml:"rag_llm"`
APIKeys *ragAPIKeysYAML `yaml:"api_keys,omitempty"`
- TokenBudget int `yaml:"token_budget,omitempty"`
- TopN int `yaml:"top_n,omitempty"`
+ TokenBudget *int `yaml:"token_budget,omitempty"`
+ TopN *int `yaml:"top_n,omitempty"`
SystemPrompt string `yaml:"system_prompt,omitempty"`
Search *ragSearchYAML `yaml:"search,omitempty"`
}
@@
type ragDefaultsYAML struct {
- TokenBudget int `yaml:"token_budget,omitempty"`
- TopN int `yaml:"top_n,omitempty"`
+ TokenBudget *int `yaml:"token_budget,omitempty"`
+ TopN *int `yaml:"top_n,omitempty"`
}
@@
var defaults *ragDefaultsYAML
if params.Config.Defaults != nil {
- d := &ragDefaultsYAML{}
- if params.Config.Defaults.TokenBudget != nil {
- d.TokenBudget = *params.Config.Defaults.TokenBudget
- }
- if params.Config.Defaults.TopN != nil {
- d.TopN = *params.Config.Defaults.TopN
- }
- if d.TokenBudget != 0 || d.TopN != 0 {
+ d := &ragDefaultsYAML{
+ TokenBudget: params.Config.Defaults.TokenBudget,
+ TopN: params.Config.Defaults.TopN,
+ }
+ if d.TokenBudget != nil || d.TopN != nil {
defaults = d
}
}
@@
if p.TokenBudget != nil {
- pipeline.TokenBudget = *p.TokenBudget
+ pipeline.TokenBudget = p.TokenBudget
}
if p.TopN != nil {
- pipeline.TopN = *p.TopN
+ pipeline.TopN = p.TopN
}Also applies to: 72-75, 100-111, 181-185
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@server/internal/orchestrator/swarm/rag_config.go` around lines 24 - 35, The
YAML model is collapsing explicit zeros because TokenBudget and TopN are plain
ints with `omitempty`; update ragPipelineYAML to use `*int` for TokenBudget and
TopN (and change the corresponding numeric fields in any related YAML structs
such as ragDatabaseYAML and the RAGDefaults YAML struct to `*int` as well), and
preserve/propagate the pointers when mapping between internal structs and these
YAML structs so nil vs. explicit 0 is preserved (look for all mappings that
build/return ragPipelineYAML, ragDatabaseYAML, and the RAGDefaults YAML structs
and assign pointer values instead of converting to plain ints).
| func GenerateRAGConfig(params *RAGConfigParams) ([]byte, error) { | ||
| pipelines := make([]ragPipelineYAML, 0, len(params.Config.Pipelines)) | ||
| for _, p := range params.Config.Pipelines { | ||
| pipelines = append(pipelines, buildRAGPipelineYAML(p, params)) | ||
| } |
There was a problem hiding this comment.
Reject mismatched same-provider API keys instead of silently preferring one.
The control-plane config model allows separate EmbeddingLLM.APIKey and RAGLLM.APIKey, but this generator overwrites the embedding path whenever both stages use the same provider. If a pipeline supplies different OpenAI/Anthropic keys for embedding vs. generation, one stage will authenticate with the wrong credential. If the server can only accept one key per provider, fail fast here (or in ParseRAGServiceConfig) when the two keys differ.
🔧 Suggested guard
import (
+ "fmt"
"path"
@@
func GenerateRAGConfig(params *RAGConfigParams) ([]byte, error) {
pipelines := make([]ragPipelineYAML, 0, len(params.Config.Pipelines))
for _, p := range params.Config.Pipelines {
- pipelines = append(pipelines, buildRAGPipelineYAML(p, params))
+ pipeline, err := buildRAGPipelineYAML(p, params)
+ if err != nil {
+ return nil, err
+ }
+ pipelines = append(pipelines, pipeline)
}
@@
-func buildRAGPipelineYAML(p database.RAGPipeline, params *RAGConfigParams) ragPipelineYAML {
+func buildRAGPipelineYAML(p database.RAGPipeline, params *RAGConfigParams) (ragPipelineYAML, error) {
@@
- apiKeys := buildRAGAPIKeysYAML(p, params.KeysDir)
+ apiKeys, err := buildRAGAPIKeysYAML(p, params.KeysDir)
+ if err != nil {
+ return ragPipelineYAML{}, err
+ }
@@
- return pipeline
+ return pipeline, nil
}
@@
-func buildRAGAPIKeysYAML(p database.RAGPipeline, keysDir string) *ragAPIKeysYAML {
+func buildRAGAPIKeysYAML(p database.RAGPipeline, keysDir string) (*ragAPIKeysYAML, error) {
+ if p.EmbeddingLLM.Provider == p.RAGLLM.Provider &&
+ p.EmbeddingLLM.APIKey != nil && p.RAGLLM.APIKey != nil &&
+ *p.EmbeddingLLM.APIKey != "" && *p.RAGLLM.APIKey != "" &&
+ *p.EmbeddingLLM.APIKey != *p.RAGLLM.APIKey {
+ return nil, fmt.Errorf(
+ "pipeline %q uses different %s API keys for embedding and rag",
+ p.Name,
+ p.RAGLLM.Provider,
+ )
+ }
+
keys := &ragAPIKeysYAML{}
@@
if keys.Anthropic == "" && keys.OpenAI == "" && keys.Voyage == "" {
- return nil
+ return nil, nil
}
- return keys
+ return keys, nil
}Also applies to: 130-160, 206-236
🤖 Prompt for AI Agents
Verify each finding against the current code and only fix it if needed.
In `@server/internal/orchestrator/swarm/rag_config.go` around lines 94 - 98,
GenerateRAGConfig currently builds pipelines by calling buildRAGPipelineYAML
without validating that per-stage credentials match; this lets a pipeline with
EmbeddingLLM.APIKey and RAGLLM.APIKey set to different values silently prefer
one key and misauthenticate. Add a guard where pipelines are iterated (in
GenerateRAGConfig, and similarly in ParseRAGServiceConfig if present) to detect
when both EmbeddingLLM.APIKey and RAGLLM.APIKey are non-empty for the same
provider (e.g., "openai", "anthropic") and differ, then return an error (fail
fast) describing the mismatched keys and include the pipeline identifier; update
buildRAGPipelineYAML callers to rely on this validation so no silent overwrite
occurs.
1 participant
Summary
This PR adds
RAGConfigResource, which generatespgedge-rag-server.yamlfrom pipeline configuration and writes it to the host data directory, completing the file-based config layer for the RAG service.Changes
rag_config.go— YAML struct definitions mirroring the RAG server'sConfigstruct andGenerateRAGConfig()generator;api_keyspathsreference bind-mounted key files at
/app/keys/{pipeline}_{embedding|rag}.keyrag_config_resource.go—RAGConfigResourcelifecycle (Create/Update/Refresh/Delete);
orchestrator.go— addsDirResource(host-side data directory) andRAGConfigResourcetogenerateRAGInstanceResources; resource chain is now:DirResource → RAGServiceUserRole + RAGServiceKeysResource → RAGConfigResourceresources.go— registersResourceTypeRAGConfigTesting
Verification:
Created a cluster
Configured a database with RAG service (single-host and multi-host)
rag_create_db.json
rag_create_multi_host_db.json
Confirmed successful database creation and generation of
pgedge-rag-server.yamlwith the expected datapgedge-rag-server.yaml
Checklist
Notes for Reviewers
Container deployment is not part of this. The config file is written correctly but no container starts yet.