Bug 11226

From cono_delia


 Hello,

 When I start piranha using this config file it goes into defunct mode, and
 I get a core dump.

 Here is my config. How can I email the core dump? Piranha in my opinion is
 not stable at all. It works when it feels like it. I can stop pulse start
 it up again and all my virtual services get killed. I sometimes have to
 reboot the machine to refresh all the buffers....

 The Config:

 primary = 172.10.20.30
 service = lvs
 rsh_command = rsh
 backup = 0.0.0.0
 heartbeat = 1
 heartbeat_port = 539
 keepalive = 6
 deadtime = 18
 network = nat
 nat_router = 192.168.1.254 eth1:0
 virtual vs1 {
      active = 1
      address = 172.10.20.254 eth0:1
      port = 443
      persistent = 1800
      load_monitor = ruptime
      scheduler = wlc
      protocol = tcp
      persistent = 1800
      timeout = 60
      reentry = 5
      server rs1 {
          address = 192.168.1.10
          active = 1
          weight = 100
      }
 }
 virtual vs2 {
      active = 1
      address = 172.10.20.254 eth0:1
      port = 25
      load_monitor = ruptime
      scheduler = wlc
      protocol = tcp
      timeout = 6
      reentry = 15
      server rs1 {
          address = 192.168.1.12
          active = 1
          weight = 100
      }
 }
 virtual vs3 {
      active = 1
      address = 172.10.20.254 eth0:1
      port = 80
      send = "GET / HTTP/1.0\r\n\r\n"
      expect = "HTTP"
      load_monitor = ruptime
      scheduler = wlc
      protocol = tcp
      timeout = 6
      reentry = 15
      server rs1 {
          address = 192.168.1.10
          active = 1
          weight = 100
      }
      server rs2 {
          address = 192.168.1.25
          active = 1
          weight = 100
      }
 }



 Thanks, Cono

 ------- Additional Comments From cono_delia  04/18/00 00:00
-------
 Created an attachment (id=205)
 this is the core dump produced when I start Piranha


 ------- Additional Comments From kbarrett  04/18/00 11:04
-------
 My attempts to read your core dump file appear to fail. Did you
 indicate "binary file" when you attached it?

 I have not seen anything like extreme behaviors you describe. I
 can tell you that Piranha LVS has existed since 6.0., and has
 gone through a lot of testing.

 There is no config file syntax checking in piranha, so it is easy
 to produce a config file that will result in a core dump if you
 edit the file manually instead of using the GUI interface (this
 is why Red Hat will not support config files unless they are
 created by the GUI). If one of the nanny processes or the lvs
 process core dumnped, you'd see a defunc'd process.

 I may be wrong, but it doesn't look like your config file was
 created by the GUI. You have set the backup IP address to
 0.0.0.0, but have no (required) parameter "backup_active = 0".
 However I don't see anything else obviously wrong with the file.

 Try the following:

 1. Run the web GUI, have it load your config file, and go through
    each screen and touch each field to ensure that it gets written
    out correctly, then save the file.

 2. Before starting piranha, check that ip_forwarding and defrag are
    set correctly by makeing sure that /proc/sys/net/ipv4/ip_forward
    and ip_always_defrag have "1" in them. If they don't, it means that
    your system startup is not setting these files correctly. Do a
    "echo >/proc/sys/... 1" to each file before starting piranha.

    Also make sure ipchains has been set correctly.

 3. Look at /var/log/messages to see WHY a program might exit if it
    did not core dump.

 4. If the problem appears to be piranha declaring your functioning
    services as disfunctional, try removing the "send" and "expect"
    strings and  see if that affects the problem.

 5. Use ipvsadm to display the status of your real servers while
    piranha is running.

 If you are still having problems, we would need to know more.
 For example; WHICH program is core dumping. What do your ipchains
 entries look like? Can you telnet from the NAT router to your
 real server's service port (443 in your case)? If telnet cannot
 reach it, then piranha won't be able to either.


 ------- Additional Comments From mcm  04/19/00 07:00 -------
 This bug can cause segfault here is a quick fix.

 --- lvs.c.old   Wed Apr 19 03:54:56 2000
 +++ lvs.c       Wed Apr 19 03:59:31 2000
 @@ -504,7 +504,7 @@

      if (vserver->persistent > 0 ) {
          *arg++ = (char *) "-p";
 -        (void) sprintf(*arg++, "%d", vserver->persistent);
 +        (void) asprintf(arg++, "%d", vserver->persistent);

          if (vserver->pmask.s_addr) {
              pmask = inet_ntoa(vserver->pmask);


 ------- Additional Comments From kbarrett  04/19/00 11:11
-------
 Re: mcm

 This may not be a bug, and may not have anything to do with this specific
 bugzilla report. The asprintf function is not a a function that can be
used by
 piranha, and is not documented in linux. Red Hat cannot recommend this
patch at
 this time. I am exchanging email with the user, and if the situation
changes I
 will log the results here.







 ------- Additional Comments From kbarrett  04/19/00 11:53
-------
 More followup...

 After an email exchange, mcm is sending me a config file that can cause a
 segfault in his indicated code area. When I receive and test this, I'll
post an
 update and/or patch. asprintf() is a GNU extention provided in glibc and
not
 well documented. It can be used if a dependancy on glibc is acceptable.
Piranha
 has no such dependancy currently.


 ------- Additional Comments From cono_delia  04/20/00 11:58
-------
 I have followed your advice and used the GUI only to create the lvs.cf
file.
 Piranha seems to be working fine now except for a few messages I get every
hour.
 For some reason I'm getting timeouts every so often on all of my real
servers.
 I also recieved a lvs1 pumpd[268] renewed lease for interface eth0. From
the
 rotuer, I tried to telnet to the ports like you suggested, I get a connect
but
 nothing happens after that.

 Cono.

 ------- Additional Comments From kbarrett  04/25/00 10:43
-------
 >Piranha seems to be working fine now except for a few messages I get
 >every hour. For some reason I'm getting timeouts every so often on
 > all of my real servers.

 If you suspect this is a performance issue, you can increase the value
 of your virtual service "timeout =" value (be sure to also increase the
 reentry timer to a multiple of the timeout value).

 >I also recieved a lvs1 pumpd[268] renewed lease for interface eth0.

 Hummm.  Sounds like your timeout may be network related rather than
 piranha itself.

 >I tried to telnet to the ports like you suggested, I get a connect
 >but nothing happens after that.

 telnetting to port 443 should just connect like you describe.
 telnetting to port 25 should display a banner and then appear dead.