[
https://issues.apache.org/jira/browse/KUDU-3749?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Gabriella Lotz updated KUDU-3749:
---------------------------------
Description:
ksck, auto_rebalancer, and PartitionKey::DebugString() hex‑encode range/hash
keys by streaming raw bytes into ostringstream. If char is signed and the range
key contains bytes ≥ 0x80, the current
static_cast<uint16_t>(range_key_begin[i]) will sign‑extend, producing incorrect
hex output (e.g., ff becomes ffff). This is a correctness issue in the encoded
representation and can lead to inconsistent range tags between components.
Proposed fix: Use an unsigned byte cast before widening in both places, e.g.:
static_cast<uint16_t>(static_cast<uint8_t>(range_key_begin[i]))
or an equivalent helper to ensure 0–255 semantics.
Locations:
src/kudu/tools/ksck.cc (range_key_begin formatting)
src/kudu/master/auto_rebalancer.cc (range_key_begin formatting)
src/kudu/common/partition.cc (PartitionKey::DebugString())
was:
Both ksck and auto_rebalancer hex‑encode range_key_begin by streaming raw bytes
into ostringstream. If char is signed and the range key contains bytes ≥ 0x80,
the current static_cast<uint16_t>(range_key_begin[i]) will sign‑extend,
producing incorrect hex output (e.g., ff becomes ffff). This is a correctness
issue in the encoded representation and can lead to inconsistent range tags
between components.
Proposed fix: Use an unsigned byte cast before widening in both places, e.g.:
static_cast<uint16_t>(static_cast<uint8_t>(range_key_begin[i]))
or an equivalent helper to ensure 0–255 semantics.
Locations:
src/kudu/tools/ksck.cc (range_key_begin formatting)
src/kudu/master/auto_rebalancer.cc (range_key_begin formatting)
> Fix range_key_begin hex encoding to avoid signed-char sign extension
> --------------------------------------------------------------------
>
> Key: KUDU-3749
> URL: https://issues.apache.org/jira/browse/KUDU-3749
> Project: Kudu
> Issue Type: Bug
> Reporter: Gabriella Lotz
> Assignee: Gabriella Lotz
> Priority: Major
>
> ksck, auto_rebalancer, and PartitionKey::DebugString() hex‑encode range/hash
> keys by streaming raw bytes into ostringstream. If char is signed and the
> range key contains bytes ≥ 0x80, the current
> static_cast<uint16_t>(range_key_begin[i]) will sign‑extend, producing
> incorrect hex output (e.g., ff becomes ffff). This is a correctness issue in
> the encoded representation and can lead to inconsistent range tags between
> components.
> Proposed fix: Use an unsigned byte cast before widening in both places, e.g.:
> static_cast<uint16_t>(static_cast<uint8_t>(range_key_begin[i]))
> or an equivalent helper to ensure 0–255 semantics.
> Locations:
> src/kudu/tools/ksck.cc (range_key_begin formatting)
> src/kudu/master/auto_rebalancer.cc (range_key_begin formatting)
> src/kudu/common/partition.cc (PartitionKey::DebugString())
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
