File descriptor leaks in V5.2.6 under Linux

Message boards : BOINC client : File descriptor leaks in V5.2.6 under Linux
Message board moderation

To post messages, you must log in.

AuthorMessage
Hermes

Send message
Joined: 5 Nov 05
Posts: 8
Germany
Message 828 - Posted: 6 Nov 2005, 0:41:04 UTC

The CoreClients forgets to close the temporary files created by mkstemp in client/http_curl.C
"lsof" wrote:
COMMAND PID USER FD TYPE DEVICE SIZE NODE NAME
boinc 11614 stephan cwd DIR 253,1 12288 962414 /home/stephan/boinc
boinc 11614 stephan rtd DIR 253,0 4096 2 /
boinc 11614 stephan txt REG 253,1 1643648 959133 /home/stephan/boinc/boinc_5.2.6
boinc 11614 stephan mem REG 253,0 112236 1534091 /lib/ld-2.3.6.so
boinc 11614 stephan mem REG 253,0 1525032 1534100 /lib/tls/libc-2.3.6.so
boinc 11614 stephan mem REG 253,0 213992 1534104 /lib/tls/libm-2.3.6.so
boinc 11614 stephan mem REG 253,0 16800 1534116 /lib/libdl-2.3.6.so
boinc 11614 stephan mem REG 253,0 107832 1534131 /lib/tls/libpthread-2.3.6.so
boinc 11614 stephan mem REG 253,0 81184 1534211 /lib/libresolv-2.3.6.so
boinc 11614 stephan mem REG 0,0 0 [heap] (stat: No such file or directory)
boinc 11614 stephan mem REG 253,0 22584 1534360 /lib/libnss_dns-2.3.6.so
boinc 11614 stephan mem REG 253,0 47380 1534371 /lib/libnss_files-2.3.6.so
boinc 11614 stephan DEL REG 0,7 3670018 /SYSV00013d8a
boinc 11614 stephan DEL REG 0,7 3768323 /SYSV01013786
boinc 11614 stephan 0u CHR 136,5 7 /dev/pts/5
boinc 11614 stephan 1w REG 253,1 4804 959354 /home/stephan/boinc/stdoutdae.txt
boinc 11614 stephan 2w REG 253,1 615411 959178 /home/stephan/boinc/stderrdae.txt
boinc 11614 stephan 3wW REG 253,1 0 962422 /home/stephan/boinc/lockfile
boinc 11614 stephan 4r DIR 253,1 4096 992700 /home/stephan/boinc/slots
boinc 11614 stephan 5u IPv4 7292107 TCP *:1043 (LISTEN)
boinc 11614 stephan 6u IPv4 7292163 TCP diana.localdomain:1043->raptor.localdomain:1457 (ESTABLISHED)
boinc 11614 stephan 7u REG 253,1 93 959373 /home/stephan/boinc/blcwbgKLb (deleted)
boinc 11614 stephan 8u REG 253,1 64 959377 /home/stephan/boinc/blcBgIgCc (deleted)
boinc 11614 stephan 9u REG 253,1 93 959471 /home/stephan/boinc/blcQ7M5Ax (deleted)
boinc 11614 stephan 10u REG 253,1 64 959460 /home/stephan/boinc/blcTw2mHT (deleted)

Furthermore these file descriptors, as well as any open internet connections, are inherited by the applications. e.g. rosetta from Rosetta@home
"lsof" wrote:
COMMAND PID USER FD TYPE DEVICE SIZE NODE NAME
rosetta_4 28611 stephan cwd DIR 253,1 4096 280606 /home/stephan/boinc/slots/1
rosetta_4 28611 stephan rtd DIR 253,0 4096 2 /
rosetta_4 28611 stephan txt REG 253,1 7749408 1105410 /home/stephan/boinc/projects/boinc.bakerlab.org_rosetta/rosetta_4.78_i686-pc-linux-gnu
rosetta_4 28611 stephan mem REG 0,0 0 [heap] (stat: No such file or directory)
rosetta_4 28611 stephan DEL REG 0,7 3768323 /SYSV01013786
rosetta_4 28611 stephan 0u CHR 136,5 7 /dev/pts/5
rosetta_4 28611 stephan 1w REG 253,1 109724 276398 /home/stephan/boinc/slots/1/stdout.txt
rosetta_4 28611 stephan 2w REG 253,1 103 276397 /home/stephan/boinc/slots/1/stderr.txt
rosetta_4 28611 stephan 3w REG 253,1 0 962422 /home/stephan/boinc/lockfile
rosetta_4 28611 stephan 4wW REG 253,1 0 276399 /home/stephan/boinc/slots/1/boinc_lockfile
rosetta_4 28611 stephan 5u IPv4 7292107 TCP *:1043 (LISTEN)
rosetta_4 28611 stephan 6u IPv4 7292163 TCP diana.localdomain:1043->raptor.localdomain:1457 (ESTABLISHED)
rosetta_4 28611 stephan 7u REG 253,1 93 959373 /home/stephan/boinc/blcwbgKLb (deleted)
rosetta_4 28611 stephan 8u REG 253,1 64 959377 /home/stephan/boinc/blcBgIgCc (deleted)
rosetta_4 28611 stephan 9r FIFO 0,5 7604148 pipe
rosetta_4 28611 stephan 10w FIFO 0,5 7604148 pipe

ID: 828 · Report as offensive
Deborah Goldsmith

Send message
Joined: 7 Nov 05
Posts: 8
United States
Message 829 - Posted: 7 Nov 2005, 0:15:22 UTC

It happens on any non-Windows platform. I already filed a BoincZilla bug (#410).
ID: 829 · Report as offensive
Sebastian Masch

Send message
Joined: 29 Aug 05
Posts: 9
Message 835 - Posted: 7 Nov 2005, 11:13:13 UTC

It has been fixed already around 3 days ago.

http://setiathome2.ssl.berkeley.edu/cgi-bin/cvsweb.cgi/boinc/client/http_curl.C
ID: 835 · Report as offensive
Rom Walton
Project developer
Avatar

Send message
Joined: 26 Aug 05
Posts: 164
Message 853 - Posted: 8 Nov 2005, 19:17:02 UTC

This bug has been fixed in build 5.2.7

Thanks for the info.

----- Rom
BOINC Development Team, U.C. Berkeley
My Blog
ID: 853 · Report as offensive
Deborah Goldsmith

Send message
Joined: 7 Nov 05
Posts: 8
United States
Message 862 - Posted: 9 Nov 2005, 2:22:20 UTC

This happens for Mac OS X as well; any estimate on when we might see 5.2.7 for Mac OS X?
ID: 862 · Report as offensive
Hermes

Send message
Joined: 5 Nov 05
Posts: 8
Germany
Message 903 - Posted: 12 Nov 2005, 14:17:43 UTC - in response to Message 853.  
Last modified: 12 Nov 2005, 14:18:30 UTC

This bug has been fixed in build 5.2.7

Thanks for the info.


Not quite yet. The temporary file are gone, but the applications still have file descriptors of the network connections open:

COMMAND PID USER FD TYPE DEVICE SIZE NODE NAME
boinc 30847 stephan 0u CHR 136,5 7 /dev/pts/5
boinc 30847 stephan 1w REG 253,1 132274 959354 /home/stephan/boinc/stdoutdae.txt
boinc 30847 stephan 2w REG 253,1 617891 959178 /home/stephan/boinc/stderrdae.txt
boinc 30847 stephan 3wW REG 253,1 0 962422 /home/stephan/boinc/lockfile
boinc 30847 stephan 4r DIR 253,1 4096 992700 /home/stephan/boinc/slots
boinc 30847 stephan 5u IPv4 14986048 TCP *:1043 (LISTEN)
boinc 30847 stephan 6u IPv4 15685671 TCP diana.localdomain:1043->raptor.localdomain:1052 (ESTABLISHED)

rosetta_4 13649 stephan 0u CHR 136,5 7 /dev/pts/5
rosetta_4 13649 stephan 1w REG 253,1 214408 81329 /home/stephan/boinc/slots/0/stdout.txt
rosetta_4 13649 stephan 2w REG 253,1 102 81328 /home/stephan/boinc/slots/0/stderr.txt
rosetta_4 13649 stephan 3w REG 253,1 0 962422 /home/stephan/boinc/lockfile
rosetta_4 13649 stephan 4wW REG 253,1 0 81330 /home/stephan/boinc/slots/0/boinc_lockfile
rosetta_4 13649 stephan 5u IPv4 14986048 TCP *:1043 (LISTEN)
rosetta_4 13649 stephan 6u IPv4 15685671 TCP diana.localdomain:1043->raptor.localdomain:1052 (ESTABLISHED)
rosetta_4 13649 stephan 7r FIFO 0,5 18584121 pipe
rosetta_4 13649 stephan 8w FIFO 0,5 18584121 pipe

sixtrack_ 31065 stephan 0u CHR 136,5 7 /dev/pts/5
sixtrack_ 31065 stephan 1w REG 253,1 132274 959354 /home/stephan/boinc/stdoutdae.txt
sixtrack_ 31065 stephan 2w REG 253,1 18 276366 /home/stephan/boinc/slots/1/stderr.txt
sixtrack_ 31065 stephan 3w REG 253,1 0 962422 /home/stephan/boinc/lockfile
sixtrack_ 31065 stephan 4wW REG 253,1 0 276367 /home/stephan/boinc/slots/1/boinc_lockfile
sixtrack_ 31065 stephan 5u IPv4 14986048 TCP *:1043 (LISTEN)
sixtrack_ 31065 stephan 6u IPv4 15685671 TCP diana.localdomain:1043->raptor.localdomain:1052 (ESTABLISHED)

ID: 903 · Report as offensive
Deborah Goldsmith

Send message
Joined: 7 Nov 05
Posts: 8
United States
Message 961 - Posted: 18 Nov 2005, 1:11:22 UTC

I'm not seeing this problem on Mac OS X with 5.2.5. Only the temp files are being left open.
ID: 961 · Report as offensive
Deborah Goldsmith

Send message
Joined: 7 Nov 05
Posts: 8
United States
Message 983 - Posted: 21 Nov 2005, 15:50:25 UTC

Any news on 5.2.7+ for Mac OS X? I have to manually restart boinc on my dual G5 every two days to keep from hitting this bug.
ID: 983 · Report as offensive
Hermes

Send message
Joined: 5 Nov 05
Posts: 8
Germany
Message 1800 - Posted: 12 Dec 2005, 0:00:15 UTC - in response to Message 903.  
Last modified: 12 Dec 2005, 0:00:39 UTC

Boinc 5.2.13 still passes the network connections on to the applications:
COMMAND  PID    USER   FD   TYPE   DEVICE    SIZE    NODE NAME
boinc   2988 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
boinc   2988 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
mfoldB125 15761 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
mfoldB125 15761 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
search_1. 21360 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
search_1. 21360 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
rosetta_4 21948 stephan    5u  IPv4     9663              TCP *:1043 (LISTEN)
rosetta_4 21948 stephan    6u  IPv4 27864059              TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)

ID: 1800 · Report as offensive
Metod, S56RKO

Send message
Joined: 9 Sep 05
Posts: 128
Slovenia
Message 1814 - Posted: 12 Dec 2005, 14:48:34 UTC - in response to Message 1800.  

Boinc 5.2.13 still passes the network connections on to the applications:
COMMAND  PID    USER   FD   TYPE   DEVICE    SIZE    NODE NAME
boinc   2988 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
boinc   2988 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
mfoldB125 15761 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
mfoldB125 15761 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
search_1. 21360 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
search_1. 21360 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
rosetta_4 21948 stephan    5u  IPv4     9663              TCP *:1043 (LISTEN)
rosetta_4 21948 stephan    6u  IPv4 27864059              TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)


An idea: what if this network connection passing is because of displaying of graphics? I've built my own BOINC CC (also 5.2.13) without graphics and I don't see any network connection open by scientiffic applications, only by BOINC CC itself.
Metod ...
ID: 1814 · Report as offensive
Hermes

Send message
Joined: 5 Nov 05
Posts: 8
Germany
Message 2004 - Posted: 16 Dec 2005, 16:53:52 UTC - in response to Message 1814.  

Boinc 5.2.13 still passes the network connections on to the applications:
COMMAND  PID    USER   FD   TYPE   DEVICE    SIZE    NODE NAME
boinc   2988 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
boinc   2988 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
mfoldB125 15761 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
mfoldB125 15761 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
search_1. 21360 stephan    5u  IPv4     9663             TCP *:1043 (LISTEN)
search_1. 21360 stephan    6u  IPv4 27864059             TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)
rosetta_4 21948 stephan    5u  IPv4     9663              TCP *:1043 (LISTEN)
rosetta_4 21948 stephan    6u  IPv4 27864059              TCP diana.localdomain:1043->raptor.localdomain:2632 (ESTABLISHED)


An idea: what if this network connection passing is because of displaying of graphics? I've built my own BOINC CC (also 5.2.13) without graphics and I don't see any network connection open by scientiffic applications, only by BOINC CC itself.


I don't have any graphical display on that machine.
Pleas note that there is only one open internet connection. netstat -p says the socket is connected tho the boinc CC. But lsof displays all processes, that have access to the connection. All applikation get a copy of the file descriptors open at the time of start and could send or receive messages. And they could accept new incoming connections.
ID: 2004 · Report as offensive

Message boards : BOINC client : File descriptor leaks in V5.2.6 under Linux

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.