[Bug]: finishFlow mutation returns "flow not found" when provider name was renamed - flow unloaded from in-memory map at startup

[psyray]

Affected Component

Other (please specify in the description)
Backend / Flow Controller

Describe the bug

When a model provider is renamed (e.g. Ollama cloud GLM -> OllamaC GLM), all flows that were created with the old provider name fail to load at service startup. The LoadFlowWorker function in pkg/controller/flow.go calls fwc.provs.LoadFlowProvider(), which returns an error because the provider name no longer exists. The flow is skipped and never added to the fc.flows in-memory map (pkg/controller/flows.go).

As a result, any subsequent GraphQL mutation or REST API call that looks up the flow in the in-memory map - finishFlow, stopFlow, putUserInput, callAssistant, PUT /flows/{id} with action: finish - returns ErrFlowNotFound ("flow not found"), even though the flow exists in the database with status waiting.

The flow is therefore permanently stuck: it cannot be finished, stopped, or resumed through the UI or API. Its Docker container keeps running as an orphan.

Expected behavior:

• Flows with a missing/renamed provider should still be loadable so they can be finished, deleted, or have their provider updated.
• Alternatively, FinishFlow (and similar mutations) should fall back to loading the flow from DB when it is not found in the in-memory map, so users can at least clean up stuck flows.
• At minimum, the error message should distinguish between "flow does not exist in DB" and "flow exists but failed to load" to aid debugging.

Actual behavior:

• finishFlow returns "flow not found" for flows that exist in DB but are not in the in-memory map due to a provider name mismatch.
• PUT /flows/{id} with action: finish returns HTTP 500 "internal server error".
• The UI "Finish" button appears to do nothing (error is swallowed or shown as generic failure).

Steps to Reproduce

1. Create several flows using a provider named Ollama cloud GLM.
2. Rename or delete that provider (e.g. rename to OllamaC GLM).
3. Restart the PentAGI service.
4. Observe the startup logs: failed to load flow X: failed to get flow provider: provider 'Ollama cloud GLM' not found for each affected flow.
5. In the UI, click "Finish" on any of the affected flows (e.g. flow ID 85).
6. The GraphQL mutation finishFlow returns: {"errors":[{"message":"flow not found","path":["finishFlow"]}],"data":null}.
7. The REST API PUT /flows/85 with {"action":"finish"} returns: {"code":"Internal","msg":"internal server error","status":"error"}.

Root cause analysis (from source code):

In pkg/controller/flows.go, FinishFlow() looks up the flow in fc.flows map:

func (fc *flowController) FinishFlow(ctx context.Context, flowID int64) error {
    fc.mx.Lock()
    defer fc.mx.Unlock()
    flow, ok := fc.flows[flowID]
    if !ok {
        return ErrFlowNotFound  // ← fails here because flow was never loaded
    }
    // ...
}

In pkg/controller/flow.go, LoadFlowWorker() fails when the provider is missing:

// LoadFlowWorker only loads flows with status Running or Waiting
// It calls fwc.provs.LoadFlowProvider() which fails if the provider name
// stored in DB no longer matches any configured provider.

In pkg/controller/flows.go, LoadFlows() silently skips failed flows:

for _, flow := range flows {
    fw, err := LoadFlowWorker(ctx, flow, ...)
    if err != nil {
        if errors.Is(err, ErrNothingToLoad) {
            continue
        }
        logrus.WithContext(ctx).WithError(err).Errorf("failed to load flow %d", flow.ID)
        continue  // ← flow is skipped, never added to fc.flows map
    }
    fc.flows[flow.ID] = fw
}

Affected flows in our instance: 15 flows (IDs: 56, 58, 59, 60, 61, 65, 66, 78, 82, 83, 84, 85, 87) are stuck with two missing providers:

• Ollama cloud GLM (12 flows) — renamed to OllamaC GLM
• Ollama cloud deepseek (1 flow, ID 56) — provider removed entirely

Workaround: Manually updating the model_provider_name column in the flows database table to match an existing provider name, then restarting the service, allows the flows to load and be finished normally.

System Configuration

• PentAGI Version: v2.1.0-879e87c (image vxcontrol/pentagi:latest)
• Deployment Type: Docker Compose
• Model Providers: 5 configured (including renamed providers: OllamaC GLM formerly Ollama cloud GLM, OllamaC DS formerly Ollama cloud deepseek)
• Database: PostgreSQL with pgvector
• Instance stats: 27 flows, 24 tasks, 136 subtasks, 18 assistants

Logs and Artifacts

Docker logs (docker logs pentagi) at startup show:

    time="..." level=error msg="failed to load flow 85: failed to get flow provider: provider 'Ollama cloud GLM' not found"
    time="..." level=error msg="failed to load flow 84: failed to get flow provider: provider 'Ollama cloud GLM' not found"
    time="..." level=error msg="failed to load flow 83: failed to get flow provider: provider 'Ollama cloud GLM' not found"
    time="..." level=error msg="failed to load flow 56: failed to get flow provider: provider 'Ollama cloud deepseek' not found"
    (repeated for each affected flow ID: 56, 58, 59, 60, 61, 65, 66, 78, 82, 83, 84, 85, 87)

GraphQL response from finishFlow mutation:

    {"errors":[{"message":"flow not found","path":["finishFlow"]}],"data":null}

REST API response from PUT /flows/85 with {"action":"finish"}:

    {"code":"Internal","msg":"internal server error","status":"error"}

Screenshots or Recordings

No response

Verification

• I have checked that this issue hasn't been already reported
• I have provided all relevant configuration files (with sensitive data removed)
• I have included relevant logs and error messages
• I am running the latest version of PentAGI