Bug 762985 (GLUSTER-1253) - Seeing traffic on not involved hosts..
Summary: Seeing traffic on not involved hosts..
Keywords:
Status: CLOSED WORKSFORME
Alias: GLUSTER-1253
Product: GlusterFS
Classification: Community
Component: transport
Version: 3.0.5
Hardware: All
OS: Linux
low
high
Target Milestone: ---
Assignee: Raghavendra G
QA Contact:
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2010-07-30 07:55 UTC by Matthias
Modified: 2011-06-22 09:49 UTC (History)
3 users (show)

Fixed In Version:
Doc Type: Bug Fix
Doc Text:
Clone Of:
Environment:
Last Closed:
Regression: RTNR
Mount Type: fuse
Documentation: DNR
CRM:
Verified Versions:


Attachments (Terms of Use)

Description Matthias 2010-07-30 05:09:52 UTC
Edit: 
Storage 1 is running glusterfs 3.0.3 built on Mar 14 2010 04:05:39
Storage 2 is running glusterfs 3.0.3 built on Mar 14 2010 04:05:39

Clients srv132,133,134 useing glusterfs 3.0.5 built on Oct 10 2010 00:00:29

Comment 1 Matthias 2010-07-30 07:55:47 UTC
Hi there,

i got a very strange issue and i cannot determine the problem.

I have 2 Storages running, both 3.0.5 Version

When i use a client on storage 1 i see some traffic on an node from storage2 + some traffic on an completly uninvolved node.


My network setup is like this pic

http://yfrog.com/f/eastoragexj/

http://img401.imageshack.us/gal.php?g=bse232eth0.png

Explanation:
==============
Storage 1 consists of:
Chassis 1:
node1, node2, node3, node4 -> unify 13TB

Storage 2 consists of:
srv231, srv232, srv233, srv234 -> unify 73TB

Clients on Chassis 2 (light orange blades)
srv132, srv133, srv134
mountpoints to storage1 and storage2
-----
srv132,133,134 reading data from storage 1 with 64 threads
storage2 is not used currently -> no jobs accessing storage2

Independend client (no glusterfs mount points or server)
node6 on chassis 1 sees traffic on eth1

Network interfaces distribution:
chassis 1
node1,2,3,4 eth0 30.11 , 30.12, 30.13, 30.14
node1,2,3,4 eth1 60.11 , 60.12, 60.13, 60.14 (should be for interaction interface for the bricks, clients should connect to this interface as well)

chassis 2
srv132,133,134 eth0 30.132, 30.133, 30.134 
srv132,133,134 eth1 60.132, 60.133, 60.134 (client interface for glfs)

chassis 1
node6 eth0 30.16
node6 eth1 60.16

orange interfaces always represents eth0
red interfaces always represents eth1


configs storage 1 http://pastebin.com/m2DDEKPG
configs storage 2 http://pastebin.com/RyXgJegT

Comment 2 Matthias 2010-12-27 08:44:29 UTC
Hi, 

it occurs again. I am not sure if its an glfsd bug or something with the network.

Lets see the details from tcpdump:
===========================

[12:31:12-root@srv14 ~]# ifconfig eth0 && ifconfig eth1
eth0      Link encap:Ethernet  Hardware Adresse 00:15:C5:FD:42:60
          inet Adresse:192.168.30.14  Bcast:192.168.30.255  Maske:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:10648496853 errors:0 dropped:0 overruns:0 frame:0
          TX packets:1109021748 errors:0 dropped:0 overruns:0 carrier:0
          Kollisionen:0 Sendewarteschlangenlänge:0
          RX bytes:12024592926561 (10.9 TiB)  TX bytes:4033570961141 (3.6 TiB)

eth1      Link encap:Ethernet  Hardware Adresse 00:15:C5:FD:42:62
          inet Adresse:192.168.60.14  Bcast:192.168.60.255  Maske:255.255.255.0
          UP BROADCAST RUNNING MULTICAST  MTU:1500  Metric:1
          RX packets:37729206203 errors:0 dropped:0 overruns:0 frame:0
          TX packets:29656361309 errors:0 dropped:0 overruns:0 carrier:0
          Kollisionen:0 Sendewarteschlangenlänge:0
          RX bytes:27337727209003 (24.8 TiB)  TX bytes:120227047524740 (109.3 TiB)

============================

[12:32:01-root@srv14 ~]# head -n 100 /tmp/glfd.dump |tail -n 50
17:51:34.830844 IP 192.168.60.89.1005 > 192.168.60.235.6996: P 7148:8504(1356) ack 281 win 1525 <nop,nop,timestamp 2366338968 3744359983>
17:51:34.831131 IP cluster1node5.intern.gatc.de.1014 > cluster1node4.intern.gatc.de.6996: P 144:288(144) ack 121 win 1525 <nop,nop,timestamp 1410381646 681749388>
17:51:34.831181 IP cluster1node5.intern.gatc.de.1015 > cluster1node4.intern.gatc.de.6996: P 144:288(144) ack 121 win 1525 <nop,nop,timestamp 1410381646 681749388>
17:51:34.831191 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1014: P 121:241(120) ack 288 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.831229 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1015: P 121:241(120) ack 288 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.831685 IP cluster1node5.intern.gatc.de.1013 > cluster1node4.intern.gatc.de.6996: P 971:1323(352) ack 8557 win 1525 <nop,nop,timestamp 1410381646 681749389>
17:51:34.831750 IP cluster1node5.intern.gatc.de.1012 > cluster1node4.intern.gatc.de.6996: P 971:1323(352) ack 10310 win 1525 <nop,nop,timestamp 1410381646 681749389>
17:51:34.831824 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1013: P 8557:8841(284) ack 1323 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.831864 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1012: P 10310:10594(284) ack 1323 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.832048 IP 192.168.60.83.1007 > 192.168.60.235.6996: . 4252:5700(1448) ack 281 win 1525 <nop,nop,timestamp 2444060750 3744359984>
17:51:34.832052 IP 192.168.60.83.1007 > 192.168.60.235.6996: . 5700:7148(1448) ack 281 win 1525 <nop,nop,timestamp 2444060750 3744359984>
17:51:34.832056 IP 192.168.60.83.1007 > 192.168.60.235.6996: P 7148:8504(1356) ack 281 win 1525 <nop,nop,timestamp 2444060750 3744359984>
17:51:34.832221 IP cluster1node5.intern.gatc.de.1013 > cluster1node4.intern.gatc.de.6996: P 1323:1546(223) ack 8841 win 1525 <nop,nop,timestamp 1410381646 681749389>
17:51:34.832252 IP cluster1node5.intern.gatc.de.1012 > cluster1node4.intern.gatc.de.6996: P 1323:1546(223) ack 10594 win 1525 <nop,nop,timestamp 1410381646 681749389>
17:51:34.832362 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1013: P 8841:9175(334) ack 1546 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.832365 IP cluster1node4.intern.gatc.de.6996 > cluster1node5.intern.gatc.de.1012: P 10594:10928(334) ack 1546 win 1525 <nop,nop,timestamp 681749389 1410381646>
17:51:34.832833 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 1653876659 win 1402 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.832870 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 2897 win 1514 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.832903 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 17377 win 1525 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.833132 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 34753 win 1525 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.833184 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 52129 win 1525 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.833342 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 69505 win 1525 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.833583 IP 192.168.60.85.1005 > 192.168.60.235.6996: . ack 98465 win 1525 <nop,nop,timestamp 2448184066 3744359995>
17:51:34.833622 IP 192.168.60.87.1005 > 192.168.60.235.6996: . ack 2399993835 win 1525 <nop,nop,timestamp 2443603783 3744359995>
17:51:34.833849 IP 192.168.60.87.1005 > 192.168.60.235.6996: . ack 1449 win 1525 <nop,nop,timestamp 2443603784 3744359995>

==================


So we can see "normal traffic" from IP cluster1node5.intern.gatc.de.1014 > cluster1node4.intern.gatc.de.6996: 
(srv14 & srv15 same server dns name would be cluster1node5...)

And we can see the traffic coming from srv85 to srv235 which is at least 1 hop away! So why this is happening?!

Does anyone have a suggestions how to investigate that more deeply?

Thx Matthias

Comment 3 Anand Avati 2010-12-29 05:57:11 UTC
This looks like something specific to your routing setup. Maybe this server as an L3 gateway for those two networks or your switch does not do L2/MAC filtering. I don't see in what possible way this is an issue with Gluster (except the traffic is Gluster traffice). Gluster uses only standard sockets for all its communication.

Avati

Comment 4 Matthias 2011-01-04 06:59:12 UTC
(Mit Bezug zu comment #3)
> This looks like something specific to your routing setup. Maybe this server as
> an L3 gateway for those two networks or your switch does not do L2/MAC
> filtering. I don't see in what possible way this is an issue with Gluster
> (except the traffic is Gluster traffice). Gluster uses only standard sockets
> for all its communication.
> 
> Avati

The wired thing is, i only see glusterfs traffic! Nothing else.
One more point is, its only ack traffic, which we can see:

 6.134719 192.168.60.13 -> 192.168.30.143 TCP 6996 > 1014 [PSH, ACK] Seq=1 Ack=1 Win=1525 Len=120 TSV=2243628531 TSER=711865572
  6.134720 192.168.60.14 -> 192.168.30.143 TCP 6996 > 1011 [PSH, ACK] Seq=1 Ack=1 Win=1525 Len=120 TSV=934533108 TSER=711865572
  6.134808 192.168.60.12 -> 192.168.30.143 TCP 6996 > 1017 [ACK] Seq=1 Ack=1 Win=1525 Len=0 TSV=934436222 TSER=711865572
  6.134810 192.168.60.13 -> 192.168.30.143 TCP 6996 > 1015 [ACK] Seq=1 Ack=1 Win=1525 Len=0 TSV=2243628531 TSER=711865572
  6.134811 192.168.60.12 -> 192.168.30.143 TCP 6996 > 1017 [PSH, ACK] Seq=1 Ack=1 Win=1525 Len=120 TSV=934436222 TSER=711865572
  6.134812 192.168.60.11 -> 192.168.30.143 TCP 6996 > 1023 [PSH, ACK] Seq=1 Ack=1 Win=1525 Len=208 TSV=931764301 TSER=711865572


60.11, 60.12, 60.13, 60.14 are the bricks server, this ack packets are running periodicly through the network, there is no data traffic on the bricks..as far i can tell. The ack packets , go to all connected clients, 60.141, 142, 143, 145 ...

So something is very strange here. If a switch is not working correctly. What i wouldn't deny thats possible, why i can only see glusterfs push/ack packets? No data packets, no other miss routed packets?

Some ideas?

Comment 5 Matthias 2011-01-04 07:01:35 UTC
(Mit Bezug zu comment #4)
> (Mit Bezug zu comment #3)
> > This looks like something specific to your routing setup. Maybe this server as
> > an L3 gateway for those two networks or your switch does not do L2/MAC
> > filtering. I don't see in what possible way this is an issue with Gluster
> > (except the traffic is Gluster traffice). Gluster uses only standard sockets
> > for all its communication.
> > 
> > Avati
> 
> The wired thing is, i only see glusterfs traffic! Nothing else.
> One more point is, its only ack traffic, which we can see:
> 
>  6.134719 192.168.60.13 -> 192.168.30.143 TCP 6996 > 1014 [PSH, ACK] Seq=1
> Ack=1 Win=1525 Len=120 TSV=2243628531 TSER=711865572
>   6.134720 192.168.60.14 -> 192.168.30.143 TCP 6996 > 1011 [PSH, ACK] Seq=1
> Ack=1 Win=1525 Len=120 TSV=934533108 TSER=711865572
>   6.134808 192.168.60.12 -> 192.168.30.143 TCP 6996 > 1017 [ACK] Seq=1 Ack=1
> Win=1525 Len=0 TSV=934436222 TSER=711865572
>   6.134810 192.168.60.13 -> 192.168.30.143 TCP 6996 > 1015 [ACK] Seq=1 Ack=1
> Win=1525 Len=0 TSV=2243628531 TSER=711865572
>   6.134811 192.168.60.12 -> 192.168.30.143 TCP 6996 > 1017 [PSH, ACK] Seq=1
> Ack=1 Win=1525 Len=120 TSV=934436222 TSER=711865572
>   6.134812 192.168.60.11 -> 192.168.30.143 TCP 6996 > 1023 [PSH, ACK] Seq=1
> Ack=1 Win=1525 Len=208 TSV=931764301 TSER=711865572
> 
> 
> 60.11, 60.12, 60.13, 60.14 are the bricks server, this ack packets are running
> periodicly through the network, there is no data traffic on the bricks..as far
> i can tell. The ack packets , go to all connected clients, 60.141, 142, 143,
> 145 ...
> 
> So something is very strange here. If a switch is not working correctly. What i
> wouldn't deny thats possible, why i can only see glusterfs push/ack packets? No
> data packets, no other miss routed packets?
> 
> Some ideas?

Update: traffic was seen on 60.131 an additional different server, which is located in another bladecenter..its getting even stranger..

Comment 6 Anand Avati 2011-01-04 09:32:44 UTC
> Update: traffic was seen on 60.131 an additional different server, which is
> located in another bladecenter..its getting even stranger..

Maybe your configuration is such that it is resulting in traffic between the two sites. As I already mentioned, glusterfs contains its entire communication within AF_INET socket API communication. Even if it wanted to, glusterfs could not be sending mis-routed traffic on the wire.

Avati

Comment 7 Amar Tumballi 2011-04-25 09:33:16 UTC
Please update the status of this bug as its been more than 6months since its filed (bug id < 2000)

Please resolve it with proper resolution if its not valid anymore. If its still valid and not critical, move it to 'enhancement' severity.

Comment 8 Matthias 2011-06-22 06:49:54 UTC
Seems to be a local network problem on Dell 5316M Switches.

We still have no idea why this is happening.

-----
The same issues occure with 3.0.7 or 3.2.x Versions of glusterfs.

I am a bit glueless. If i ran iperf on the storages i get good performance, about 1Gbit.

[09:31:08 root@srv11 glusterfs]# iperf -s
------------------------------------------------------------
Server listening on TCP port 5001
TCP window size: 85.3 KByte (default)
------------------------------------------------------------
[  4] local 192.168.30.11 port 5001 connected with 192.168.30.20 port 48902
[ ID] Interval       Transfer     Bandwidth
[  4]  0.0-10.0 sec  1.10 GBytes    941 Mbits/sec

------

But if i tranfer data, only over this switch, means internal bladechassis switch, the traffic is more than bad and i see lost connections in the logfile.

------
[2011-06-22 08:46:41] N [client-protocol.c:6329:client_setvolume_cbk] 192.168.60.14-1: Connected to 192.168.60.14:6996, attached to remote volume 'brick1'.
[2011-06-22 08:46:41] N [client-protocol.c:6329:client_setvolume_cbk] 192.168.60.14-1: Connected to 192.168.60.14:6996, attached to remote volume 'brick1'.
[2011-06-22 08:46:41] W [fuse-bridge.c:1179:fuse_err_cbk] glusterfs-fuse: 5798: FLUSH() ERR => -1 (Transport endpoint is not connected)
[2011-06-22 08:47:23] E [client-protocol.c:415:client_ping_timer_expired] 192.168.60.11-1: Server 192.168.60.11:6996 has not responded in the last 42 seconds, disconnecting.
[2011-06-22 08:47:23] E [saved-frames.c:165:saved_frames_unwind] 192.168.60.11-1: forced unwinding frame type(1) op(WRITE)
[2011-06-22 08:47:23] E [saved-frames.c:165:saved_frames_unwind] 192.168.60.11-1: forced unwinding frame type(1) op(WRITE)
[2011-06-22 08:47:23] W [fuse-bridge.c:936:fuse_setattr_cbk] glusterfs-fuse: 5806: SETATTR() /pacbio/common/test/jobs/lambda/results/accuracyHistogram.png => -1 (Transport endpoint is not connected)
[2011-06-22 08:47:23] E [saved-frames.c:165:saved_frames_unwind] 192.168.60.11-1: forced unwinding frame type(2) op(PING)
[2011-06-22 08:47:23] N [client-protocol.c:7077:notify] 192.168.60.11-1: disconnected
[2011-06-22 08:47:23] N [client-protocol.c:6329:client_setvolume_cbk] 192.168.60.11-1: Connected to 192.168.60.11:6996, attached to remote volume 'brick1'.

-----
Someone got a glue? Else this is closed for me..

M


Note You need to log in before you can comment on or make changes to this bug.