Menu
▾
▴
#2325 Handle SIGPIPE in providerMgr
pending-fixed
5
2012-01-25
2012-01-24
No
On some platform a situation may occur while chunking is enabled when a client times out before a request is completed. In this case, the provider will attempt to write the first chunk to the pipe to the httpAdapter process. Since the client died, the httpAdapter process is also gone. This write results in a SIGPIPE being sent to the provider which will cause it to quietly exit (on some platforms, SIGPIPE results in a SIGTERM to the process).
By default, providerMgr should just log this signal and let the provider continue. If desired, it would be more efficient for individual providers to handle the signal themselves and do the proper steps to abandon that request.
Patch based on CVS head.
Git patch