fix(argoCD): avoid duplication of latest deployments

[antarasevich]

Summary

Fix duplicate ArgocdSyncOperation records caused by ArgoCD exposing the same completed sync in two places in its API response: status.history[] and status.operationState.

Problem

ArgoCD includes the most recent completed sync in both:

• status.history[] — the authoritative log of all past syncs, each with deployedAt
• status.operationState — a mirror of the last operation, with startedAt

The DevLake collector feeds both into the raw table. The extractor then processes each independently, using different timestamp fields as DeploymentId:

Source	Field used	Example value	DeploymentId
history[id=30]	deployedAt	2026-04-16T09:02:16Z	1776330136
operationState	startedAt	2026-04-16T09:02:15Z	1776330135

Because startedAt and deployedAt differ by one second (the sync duration), the two records get different primary keys and both are persisted — resulting in duplicate deployment entries in DORA metrics and deployment frequency charts.

Fix

Extract the skip logic into a dedicated isRedundantOperationState(phase string) bool function and call it from inside Extract before processing any operationState row.

A Succeeded operationState is always already present in status.history[] so it is skipped. In-flight (Running, Terminating) and failure (Failed, Error) states are not yet — or may never be — written to history, so they continue to be processed normally.

// isRedundantOperationState reports whether an operationState entry should be
// skipped because it duplicates the most recent history entry.
func isRedundantOperationState(phase string) bool {
    return phase == "Succeeded"
}

// inside Extract:
isOperationState := apiOp.Phase != ""

if isOperationState && isRedundantOperationState(apiOp.Phase) {
    return nil, nil
}

Behaviour by phase

operationState phase	Already in history[]?	Action
Succeeded	yes	skip — history entry is authoritative
Running / Terminating	no	keep — only source for in-flight sync
Failed / Error	no	keep — failure may never reach history
(empty)	n/a — history entry, not operationState	keep — guard never fires

Testing

Added unit tests for isRedundantOperationState covering all phase variants. Verified against production ArgoCD API responses for multiple applications where operationState.phase = Succeeded — duplicate ArgocdSyncOperation records are eliminated after the fix.

Related

• Affects: backend/plugins/argocd/tasks/sync_operation_extractor.go
• Reproduces with: any ArgoCD application whose last sync completed successfully (i.e. the vast majority of applications in a healthy cluster)

[rbstp]

Suggested change

// *** THE FIX ***

[rbstp]

Thanks for the contribution! Can you please remove the above comment, and fix the failing test.

[antarasevich]

invalid. To be updated