Skip to content

Crater v1.4.2

Latest

Choose a tag to compare

@AkashiSensei AkashiSensei released this 16 Nov 11:25
· 3 commits to main since this release
f0d3a0b

Version 1.4.2

This release introduces an early preview of Crater-Agent with Helm deployment and model configuration support, enables Python process profiling in jobs, and fixes download and approval workflow issues.

What's new

  • Crater-Agent early preview: Adds platform guidance, basic chat, and conversation history. This is an initial testing release; capabilities and the overall experience will be improved gradually based on feedback.
  • Agent deployment: The Helm chart now supports deploying Crater-Agent as a separate internal service, with Secret-based authentication and NetworkPolicy configuration. It is disabled by default. To enable it, create a Secret containing internal-token, context-signing-key, and encryption-key, then configure agent.enabled and agent.existingSecretName. Preserve the encryption key across upgrades to keep existing personal model credentials readable.
  • Agent model configuration: Supports the platform's default OpenAI-compatible model configuration and user-specific overrides. Personal model API keys are stored encrypted, and administrators can review conversation history for operations and troubleshooting.
  • Job performance profiling: Supports sampling running Python processes with py-spy to help locate performance bottlenecks.
  • Download fixes: Completed downloads now link to resources by ID, avoiding confusion between resources with the same name. Model download pause and resume handling is also more reliable.
  • Approval workflow fix: Approval requests can now be submitted when no reviewer has been assigned yet.

版本 1.4.2

本次更新引入 Crater-Agent 初步测试版,提供配套的 Helm 部署和模型配置支持,支持作业内 Python 进程性能分析,并修复下载和审批流程中的问题。

本次更新

  • Crater-Agent 初步测试版:支持平台使用咨询、基础对话和历史会话。当前仍处于初步测试阶段,我们会根据试用反馈逐步完善功能和体验。
  • Agent 部署支持:Helm Chart 新增独立的内部 Agent 服务,配套提供基于 Secret 的服务认证和 NetworkPolicy 配置,默认不启用。启用前需创建包含 internal-token、context-signing-key 和 encryption-key 的 Secret,并配置 agent.enabled 和 agent.existingSecretName。升级时需保留原有加密密钥,确保已有个人模型凭据仍可正常读取。
  • Agent 模型配置:支持使用平台默认的 OpenAI 兼容模型配置,也支持用户配置个人模型。个人模型 API 密钥加密存储,管理员可查看历史对话,用于运维和故障排查。
  • 作业性能分析:支持使用 py-spy 对作业中运行的 Python 进程采样,帮助定位性能瓶颈。
  • 下载相关修复:已完成的下载通过资源 ID 关联目标资源,避免跳转到同名资源;同时改善模型下载暂停和恢复时的状态处理。
  • 审批流程修复:修复审批工单尚未分配审批人时可能提交失败的问题。

Full changelog / 完整变更: v1.4.1...v1.4.2