Jens Geyer created THRIFT-6226:
----------------------------------
Summary: C++: TServerSocket::listen() tries only the first
resolved address unless a retry limit is set
Key: THRIFT-6226
URL: https://issues.apache.org/jira/browse/THRIFT-6226
Project: Thrift
Issue Type: Bug
Components: C++ - Library
Reporter: Jens Geyer
The TCP bind loop in {{TServerSocket::listen()}} ({{TServerSocket.cpp}}, lines
508-545 on master) is commented "we iterate over what system gave us, picking
the first address that works", but it ends with
{code:cpp}
} while ((retries++ < retryLimit_) && (THRIFT_SLEEP_SEC(retryDelay_) == 0));
{code}
and {{retryLimit_}} is 0 after every constructor. When the first address fails
to bind, the loop ends, and the next address is never tried unless the caller
has used {{setRetryLimit()}}. A {{socket()}} failure on the first address --
{{AF_INET6}} on a host without IPv6, say -- ends it the same way, because
{{continue}} in a {{do ... while}} jumps straight to that condition.
So a host name that resolves to more than one address fails to listen whenever
the first one cannot be used, even if a later one would work. Found while
looking at THRIFT-6191.
Suggested fix: try every resolved address in each attempt, and apply the retry
limit to whole passes over the list.
_Drafted with AI assistance (Claude Opus 5); reviewed and filed by Jens Geyer._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
