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

[Bug] Upper-case hex in cloud initial rowset id is mis-parsed by BE from_hex (lossy, non-unique RowsetId)

未关闭 适合新手
  • 哪里不对？
#69,371 2 条评论 0 个 reaction 已指派 0 人 在 GitHub 查看

维护者通常 1 天内回复

还没有人认领这个 Issue。

评估

难度
2/5
预计耗时
1-3 小时
新手友好度
78/100
Issue 类型
缺陷
描述清晰度
描述清楚
活跃度
活跃
技术栈
cpp, java
领域
backend, databases

调研方向

先阅读 be/src/util/uid_util.h 和 be/src/storage/rowset_id.cpp，然后检查 issue 中提到的 from_hex() 调用处。复现大写解析的情况，并检查现有的 UID 或 rowset ID 解析单元测试。完成标准是：大写十六进制解析出的值与小写相同，且无效字符得到安全处理。

由索引模型根据 Issue 内容生成。

描述

Search before asking
  • I had searched in the issues and found no similar issues.
Version

4.1.4 (cloud mode). master is the same (351d9388d5):

  • FE fe/fe-core/src/main/java/org/apache/doris/cloud/datasource/CloudInternalCatalog.java:690-692 builds the initial [0-1] rowset id with String.format("%016X", ...) (upper-case hex).
  • BE be/src/util/uid_util.h:48-61 from_hex() only handles 0-9 and a-f; any other character is mapped with c - 'a' + 10.
  • BE be/src/storage/rowset_id.cpp:57-63 RowsetId::init(std::string_view) parses the 48-char id with from_hex().
Description

In cloud mode the FE creates the empty [0-1] rowset of every new tablet with a V2 rowset id formatted as upper-case hex:

String rowsetIdV2Str = String.format("%016X", 2L << 56)
        + String.format("%016X", uuid.getMostSignificantBits())
        + String.format("%016X", uuid.getLeastSignificantBits());

The BE parses rowset ids with from_hex(), which treats every non-digit as lower-case:

if (i >= '0' && i <= '9') {
    buf_val = i - '0';
} else {
    buf_val = i - 'a' + 10;   // 'A'..'F' -> -22..-17
}
val <<= 4;
val = val | buf_val;          // a negative nibble sets all upper bits

For 'A'..'F' the nibble is negative, so val | buf_val sets every higher bit. Any UUID half that contains an upper-case letter parses to a value like 0xffffffffffffffea. The BE then shows these rowsets as, e.g.,

0200000000000000ffffffffffffffeafffffffffffffffe

Consequences:

  1. The parse is lossy: everything before the last letter in each 16-char half is lost, so [0-1] rowset ids of different tablets can parse to the same RowsetId.
  2. RowsetId::to_string() never round-trips to the id stored in the meta-service, so any code that compares the BE-side id string with the stored one sees a mismatch for these rowsets.
  3. These ids look like version-hole rowsets in BE logs, which is confusing when diagnosing rowset reference counts.

The [0-1] rowset is empty (0 segments, 0 rows), so we have not seen data impact, but the ids are wrong and not unique.

Solution

Either side fixes it; both are one-liners:

  1. BE: make from_hex() accept upper case, e.g. else if (i >= 'A' && i <= 'F') buf_val = i - 'A' + 10; (and ideally reject anything else instead of producing a negative nibble).
  2. FE: format with %016x (lower case), matching how the BE prints rowset ids.

Doing (1) keeps existing upper-case ids already stored in the meta-service parsing correctly.

How to Reproduce?
  1. Start a cloud-mode cluster, create any table.
  2. Look at the [0-1] rowset id of a tablet on the BE (e.g. in BE logs or via the tablet meta HTTP API): it shows as 0200000000000000ffff..., while the meta-service stores 0200000000000000 + 32 upper-case hex chars.
  3. Unit level: RowsetId id; id.init("0200000000000000ABCDEF0123456789ABCDEF0123456789"); gives mi and lo with all upper bits set (e.g. 0xffffffef0123456789-style values truncated to 64 bits) instead of 0xABCDEF0123456789.
Anything Else?

from_hex() is also used for UniqueId parsing; any other caller that can receive upper-case hex is affected the same way.

Are you willing to submit PR?
  • Yes I am willing to submit a PR!
Code of Conduct
主要语言
Java
星标
16.1k
派生
4k
平均合并
2 天 17 小时
30 天内合并 PR
669

环境准备

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/doris 的其他 Issue

查看 apache/doris 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。