Bug 54365 - Connection closed by server during file transfer
Summary: Connection closed by server during file transfer
Status: NEW
Alias: None
Product: Apache httpd-2
Classification: Unclassified
Component: mpm_event (show other bugs)
Version: 2.4.3
Hardware: Sun Solaris
: P2 normal (vote)
Target Milestone: ---
Assignee: Apache HTTPD Bugs Mailing List
URL:
Keywords:
Depends on:
Blocks:
 
Reported: 2013-01-02 12:15 UTC by Jon Hadfield
Modified: 2013-01-02 12:15 UTC (History)
0 users


Attachments
Add an attachment (proposed patch, testcase, etc.)

Note You need to log in before you can comment on or make changes to this bug.
Description Jon Hadfield 2013-01-02 12:15:38 UTC 
Apache 2.4.3 - Solaris 10 sparc 64, event-mpm
APR - 1.4.6, APR util 1.4.1

Whilst downloading a file, apache is closing the connection before the transfer is complete. 

Truss output shows this as each connection is closed before the client initiates a retry and a partial response is returned:
17543/1:        read(7, " !", 1)                                = 1
17543/1:        lwp_wait(2, 0xFFFFFFFF7FFFF26C)                 = 0
17543/1:        lwp_kill(67, SIGHUP)                            = 0
17543/67:           Received signal #1, SIGHUP, in port_getn() [caught]
17543/67:             siginfo: SIGHUP pid=17543 uid=4001360 code=-1
17543/67:       port_getn(3, 0x100429558, 64, 1, 0xFFFFFFFF727FBA90) Err#4 EINTR
17543/67:       lwp_sigmask(SIG_SETMASK, 0xFFBEE007, 0x0000FFF7) = 0xFFBFFEFF [0x0000FFFF]
17543/67:       setcontext(0xFFFFFFFF727FB160)
17543/67:       port_dissociate(3, 4, 0x00000006)               = 0
17543/67:       port_dissociate(3, 4, 0x00000005)               = 0
17543/67:       close(6)                                        = 0
17543/67:       close(5)                                        = 0

Symptoms look similar to previous APR related bug: https://issues.apache.org/bugzilla/show_bug.cgi?id=42580

I'm able to reproduce by starting with StartServers 120, ThreadsPerChild 64, then starting a single request to download a large file. As apache clears down the idle threads, the active thread appears to be killed along with the idle ones.
So far unable to reproduce with StartServers 3, ThreadsPerChild 25.