Skip to main content

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[0.48.0] - 2026-02-10​

Features​

  • Configurable HTTP Timeout for MCP Calls​

    GitKnowledgeSource now supports a configurable HTTP timeout for MCP API calls via spec.mcpServer.httpTimeoutSeconds. Previously, a hardcoded 30-second timeout caused large documents (100KB+) to fail ingestion when the MCP server needed more processing time, resulting in SyncPartial warnings and incomplete syncs.

    The default timeout is now 120 seconds, which handles typical large documents without user intervention. For repositories with very large files, the timeout can be increased up to 600 seconds. A minimum of 5 seconds is enforced via CRD validation.

    spec:
      mcpServer:
        url: http://mcp-server.dot-ai.svc:3456
        httpTimeoutSeconds: 180  # optional, default 120

    See the Knowledge Source Guide for configuration details. (#49)

[0.47.0] - 2026-02-06​

Bug Fixes​

  • Fix GitKnowledgeSource Continuous Sync Loop​

    GitKnowledgeSource no longer syncs continuously in a tight loop. Previously, every status update triggered a new reconcile, causing the controller to re-clone and re-sync the repository non-stop instead of waiting for the configured schedule.

    The controller now uses GenerationChangedPredicate to only reconcile on spec changes. Scheduled syncs via cron or interval expressions continue to work as configured. (#48)

[0.46.0] - 2026-02-06​

Features​

  • GitKnowledgeSource Sync Phase Visibility​

    GitKnowledgeSource now shows a Phase column in kubectl get gks output, providing immediate feedback on sync status. Previously, the status only updated after the entire sync completed, leaving users unable to tell whether a configuration fix was picked up.

    The Phase field transitions through Syncing, Synced, and Error states. When reconciliation starts, the phase is set to Syncing and persisted immediately — before cloning or ingesting documents — so users see progress right away. On completion it moves to Synced, or Error if any issues occurred (authentication failures, clone errors, partial sync failures). (#46)

[0.45.0] - 2026-02-06​

Features​

  • GitKnowledgeSource CRD​

    Automatically sync documentation from Git repositories to the MCP knowledge base. Previously, users had to manually ingest documents or build custom tooling to keep their knowledge base current with repository changes.

    The new GitKnowledgeSource CRD provides a declarative way to specify Git repositories and file patterns for ingestion. The controller clones repositories, matches files using glob patterns (e.g., docs/**/*.md), and syncs them to MCP. Change detection ensures only modified files are processed on subsequent syncs, making it efficient for large repositories. Scheduled sync supports both cron expressions (0 3 * * *) and intervals (@every 24h), with a default of daily syncs staggered across resources to avoid thundering herd issues.

    Key capabilities include private repository support via token authentication, file size filtering with maxFileSizeBytes to skip large generated files, and detailed status reporting showing document counts, skipped files, and sync errors. The deletionPolicy field controls whether documents are removed from MCP when the CR is deleted (default: Delete) or retained for migration scenarios.

    Configure a knowledge source by creating a CR with the repository URL, branch, file patterns, and MCP server endpoint. The controller handles cloning, change detection, and cleanup automatically.

    See the Knowledge Source Guide for configuration details and examples. (#44)

[0.44.1] - 2026-01-28​

Bug Fixes​

  • Fix Infinite Reconciliation Loop When Status Exceeds etcd Size Limit

    Resolves an issue where the ResourceSyncConfig controller would enter an infinite reconciliation loop when the status object grew too large, exceeding etcd's 3MB object size limit. This caused massive CPU usage and log flooding as the controller repeatedly attempted status updates that could never succeed.

    The controller now caps the SyncErrors counter at 100,000 to prevent unbounded growth and truncates error messages to 1KB maximum. When status updates fail due to "entity too large" errors, the controller enters a 5-minute backoff period instead of retrying immediately. All status fields are sanitized before each update to ensure they fit within etcd limits.

    If you encounter this issue before upgrading, the workaround is to delete and recreate the affected ResourceSyncConfig to reset its status. (#42)

[0.44.0] - 2026-01-23​

Features​

  • Global Annotations Support in Helm Chart

    The Helm chart now supports global annotations that apply to all rendered Kubernetes resources. Previously, users couldn't add custom annotations for tools like Reloader, compliance requirements, or consistent metadata across deployments.

    Set annotations in your Helm values to apply annotations to Deployments, Pods, ServiceAccounts, ClusterRoles, and ClusterRoleBindings. For example, enabling Reloader for automatic rolling updates when ConfigMaps change:

    annotations:
      reloader.stakater.com/auto: "true"
      company.com/managed-by: "platform-team"

    Resource-specific annotations (when available) take precedence over global annotations, allowing overrides where needed. CRDs are intentionally excluded since annotation tools like Reloader watch workloads, not API schemas.

    See the Setup Guide for the full configuration reference. (#39)

[0.43.1] - 2026-01-19​

Documentation​

  • Stack Installation Notes in Feature Guides

    The Resource Sync Guide and Capability Scan Guide now include a "Stack Installation" section that clarifies these features are automatically configured when using the DevOps AI Toolkit Stack. Users who installed via the Stack can verify their configuration exists and skip the manual Quick Start steps. This helps users understand they only need to follow the manual setup if customizing the configuration or installing the controller individually. (#38)