Hacktoberfest 2026:维护者为十月标记出来的 issue,仍然开放、适合新手。 浏览 Hacktoberfest issue

Allow supplying a KmsClient instance/supplier instead of only a reflectively-instantiated class name

已关闭
#3,683 5 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

还没有人认领这个 Issue。

评估

难度
5/5
预计耗时
一周以上
新手友好度
35/100
Issue 类型
功能
描述清晰度
基本清楚
活跃度
活跃
技术栈
java
领域
backend, security

调研方向

首先阅读 KeyToolkit.getKmsClient(...) 以及现有的 CryptoFactory、DecryptionPropertiesFactory 和 EncryptionPropertiesFactory 反射路径。与维护者确认首选的 API 和存储机制,然后为提供的、没有无参构造函数的 KmsClient 增加覆盖,同时保留反射构造和 initialize(...) 的行为。

由索引模型根据 Issue 内容生成。

描述

Describe the enhancement requested

The key-tools KMS integration (org.apache.parquet.crypto.keytools) instantiates the KmsClient purely by reflection from a class name: KeyToolkit.getKmsClient(...) reads parquet.encryption.kms.client.class and calls newInstance(), requiring a public no-arg constructor, with credentials expected to arrive later via KmsClient.initialize(conf, kmsInstanceID, kmsInstanceURL, accessToken). The same reflective no-arg pattern applies to CryptoFactory / DecryptionPropertiesFactory / EncryptionPropertiesFactory via parquet.crypto.factory.class.

This works well when a KMS client is stateless and can bootstrap all of its credentials from the Configuration plus the access-token string. It does not work for clients that must be constructed with their dependencies and can't be reduced to a class name + string token, e.g.:

  • clients holding a live, credential-bearing SDK handle (federated / workload-identity credentials that aren't representable as a token string);
  • clients created and wired by a dependency-injection container;
  • in-memory / fake KMS clients used in tests, which carry per-test state and have no meaningful no-arg form.

For these, users must build a static side-channel: register the real instance in a static map keyed by a UUID written into the Configuration, point parquet.encryption.kms.client.class at a thin reflective shim that looks the instance back up in initialize(), and override parquet.encryption.kms.instance.id per instance to avoid colliding on KeyToolkit's per-(kmsInstanceID, accessToken) client cache. That's global mutable state with its own lifecycle/leak management and a one-Configuration-per-client invariant — boilerplate every such user reinvents.

Proposal. Add an opt-in, fully backward-compatible way to supply a pre-built KmsClient (or a Supplier<KmsClient> / small factory) programmatically, which KeyToolkit.getKmsClient(...) prefers over class-name reflection when present. Reflection stays the default, so existing configs are untouched. Rough shape (names TBD):

// today (still works):
conf.set("parquet.encryption.kms.client.class", "com.example.MyKmsClient");

// proposed addition:
KeyToolkit.setKmsClientFactory(conf, () -> myPreBuiltKmsClient); // or a KmsClientFactory

initialize(...) would still be invoked on the supplied instance, so credential/token plumbing is unchanged.

Scope / non-goals. No new dependencies and no vendor-specific code — this is only about how a KmsClient is provided, not which one. PropertiesDrivenCryptoFactory, the KeyMaterial format, caching, per-column keys, and key-rotation tooling are all unchanged; this only lifts the requirement that the client be reflectively no-arg constructible.

Question for maintainers. Would a change along these lines be welcome? And do you prefer (a) a Supplier<KmsClient> set on the Configuration / read-write options, or (b) a settable KmsClientFactory on KeyToolkit? Happy to implement it and open a PR (with tests using a non-no-arg client) once there's agreement on direction.

Component(s)

Core

主要语言
Java
星标
3.1k
派生
1.6k
平均合并
6 天 16 小时
30 天内合并 PR
36

贡献指南

这个仓库没有索引到贡献指南

从这里开始

  1. 先读完整个 Issue,再读项目的贡献指南。
  2. 在 Issue 下留言说明你要接手 —— 这能避免两个人做同样的事。
  3. Fork 仓库,在一个分支上完成修改。
  4. 提交 Pull Request,并在描述里引用这个 Issue 编号。

apache/parquet-java 的其他 Issue

查看 apache/parquet-java 的全部 Issue

相似的 Issue

更多 Java Issue

把新 issue 发到你的邮箱

精选适合新手参与的 GitHub issue 摘要。