Jens Geyer created THRIFT-6273:
----------------------------------
Summary: C++: TNonblockingIOThread::breakLoop(true) aborts the
whole process
Key: THRIFT-6273
URL: https://issues.apache.org/jira/browse/THRIFT-6273
Project: Thrift
Issue Type: Bug
Components: C++ - Library
Reporter: Jens Geyer
{{lib/cpp/src/thrift/server/TNonblockingServer.cpp:1501}}:
{code:cpp}
void TNonblockingIOThread::breakLoop(bool error) {
if (error) {
TOutput::instance().printf("TNonblockingServer: IO thread #%d exiting with
error.", number_);
// TODO: figure out something better to do here, but for now kill the
// whole process.
TOutput::instance().printf("TNonblockingServer: aborting process.");
::abort();
}
...
{code}
One I/O thread that cannot continue takes down the process, including any other
server or application code sharing it, with no chance for the caller to shut
down in an orderly way. The {{TODO}} in the code says as much and has been
there since the function was written.
Both call sites that pass {{true}} are in
{{TNonblockingIOThread::notifyHandler()}} in the same file -- a short read on
the notify socket ({{:1481}}) and a {{read()}} failure on it that is not
{{EWOULDBLOCK}}/{{EAGAIN}} ({{:1492}}). So this is an internal-pipe failure,
not something a peer drives; it is the design question the comment already
raises.
h2. Suggested
Decide what the failure should do instead: stop that I/O thread and let the
server report the failure through {{TServerEventHandler}}, or propagate it so
{{TNonblockingServer::serve()}} can return. Either way the {{TODO}} can then go.
_Drafted with AI assistance (Claude Opus 5); reviewed and filed by Jens Geyer._
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
