[
https://issues.apache.org/jira/browse/THRIFT-6298?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Jens Geyer resolved THRIFT-6298.
--------------------------------
Fix Version/s: 0.26.0
Assignee: Jens Geyer
Resolution: Fixed
> Rust: split_halves_must_not_clobber_each_others_timeout fails on kernels with
> HZ=250
> ------------------------------------------------------------------------------------
>
> Key: THRIFT-6298
> URL: https://issues.apache.org/jira/browse/THRIFT-6298
> Project: Thrift
> Issue Type: Bug
> Components: Rust - Library
> Reporter: Jens Geyer
> Assignee: Jens Geyer
> Priority: Minor
> Fix For: 0.26.0
>
> Time Spent: 20m
> Remaining Estimate: 0h
>
> The test
> {{transport::socket::tests::split_halves_must_not_clobber_each_others_timeout}}
> ({{lib/rs/src/transport/socket.rs}}, from THRIFT-5954) sets a read timeout
> of 250 ms and expects to read exactly that value back. Linux stores socket
> timeouts in jiffies and rounds up. On a kernel with {{CONFIG_HZ=250}}, which
> is the Debian default, the value comes back as 252 ms:
> {noformat}
> assertion `left == right` failed
> left: Some(252ms)
> right: Some(250ms)
> {noformat}
> This was seen with Rust 1.85.1 on Debian 13 (kernel 6.12). CI passes, so the
> kernels on its runners keep 250 ms exact.
> h2. Suggested
> Use timeout values that are multiples of 20 ms, which kernels with HZ 100,
> 250 and 1000 all store exactly. For example, use 240 ms instead of 250 ms;
> the test then passes on the HZ=250 machine. The test's other values, 80 ms
> and 500 ms, already qualify.
> _Drafted with AI assistance (Claude Opus 5); reviewed and filed by Jens
> Geyer._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
