Re: Problems using qmail on very large site
Hi Mark, They stay running forever and the parent process is qmail-queue. Now I have 162 defunct process. Today I will try to alter rlim_fd_cur to 64 (the number of file descriptors - Today the value is 1024. 64 was the original value) and keep the rlim_fd_max to 2048, like it is today and I will reboot the machines. I read somewhere that isn't a good idea change the default current limit of file descriptors like I did. They suggest only change the max number of file descriptor. Well, is just a guess, but I think that won't cause any damage to try ;-) Thanks for your attention ! Regards. The problem is the fact that I'm having too much defunct process. Usually I have between 350-500 process running by machine. From that, normally I have between 90-120 defunct process per machine. Do they stay forever, or do they go away? Which qmail process is the parent? I have seen Solaris 2.x systems where processes do stay around forever, but I have not seen in on, eg, FreeBSD. Regards.
Re: Problems using qmail on very large site
Yuan, For change the current and max number of file descriptors in Solaris 2.6 (in 2.7 I think is the same but I'm not sure), you need add two entries in your /etc/system file: set rlim_fd_cur=1024 set rlim_fd_max=2048 and you MUST reboot your machine in order to changes take effect. To see if everythings works, try to use the command ulimit -a before and after the modifications. Unfortanately the max number of file descriptors wouldn't be showed, just the current will (open files parameter). Hope this helps. Regards, Claudio Hi, Excuse me for asking a silly question. In Solaris 2.6 or 2.7, how do you change rlim_fd_max etc? I cannot find it in the Answer Books. Thank you very much. Yuan Hi Mark, They stay running forever and the parent process is qmail-queue. Now I have 162 defunct process. Today I will try to alter rlim_fd_cur to 64 (the number of file descriptors - Today the value is 1024. 64 was the original value) and keep the rlim_fd_max to 2048, like it is today and I will reboot the machines. I read somewhere that isn't a good idea change the default current limit of file descriptors like I did. They suggest only change the max number of file descriptor. Well, is just a guess, but I think that won't cause any damage to try ;-) Thanks for your attention ! Regards. The problem is the fact that I'm having too much defunct process. Usually I have between 350-500 process running by machine. From that, normally I have between 90-120 defunct process per machine. Do they stay forever, or do they go away? Which qmail process is the parent? I have seen Solaris 2.x systems where processes do stay around forever, but I have not seen in on, eg, FreeBSD. Regards.
RE: Problems using qmail on very large site
Claudio, I looked at my record just now. I have a Sun Ultra 5 running Solaris 2.7. When I increased the number of concurrent remote processes to 50, the log file shows "cannot open pipe" error for a lot of the processes. Is this caused by rlim_fd_max too small? Sincerely, Yuan Yuan, For change the current and max number of file descriptors in Solaris 2.6 (in 2.7 I think is the same but I'm not sure), you need add two entries in your /etc/system file: set rlim_fd_cur=1024 set rlim_fd_max=2048 and you MUST reboot your machine in order to changes take effect. To see if everythings works, try to use the command ulimit -a before and after the modifications. Unfortanately the max number of file descriptors wouldn't be showed, just the current will (open files parameter). Hope this helps. Regards, Claudio
Re: Problems using qmail on very large site
Yuan, In my case, I changed both (rlim_fd_max and rlim_fd_cur to 2048 and 1024 respectively). After reboot my machines I don't received more messages like that. By the way, my concurrency remote is set to 255. So, I believe that for your case (50) this set (2048) should work fine too. I suggest that you try change just the rlim_fd_max, because I think that change rlim_fd_cur is unnecessary or, even (I still couldn't try decrease it like I said that I will), cause that problem that originate all this discussion: a lot of defunct process running in my machine ! But in my case, like I said, I'm using Solaris 2.6 and not 2.7 like you Regards, Claudio Is this caused by rlim_fd_max too small? Sincerely, Yuan Yuan, For change the current and max number of file descriptors in Solaris 2.6 (in 2.7 I think is the same but I'm not sure), you need add two entries in your /etc/system file: set rlim_fd_cur=1024 set rlim_fd_max=2048 and you MUST reboot your machine in order to changes take effect. To see if everythings works, try to use the command ulimit -a before and after the modifications. Unfortanately the max number of file descriptors wouldn't be showed, just the current will (open files parameter). Hope this helps. Regards, Claudio
Problems using qmail on very large site
Hi everybody, I installed qmail (v. 1.03 running on Solaris 2.6 for x86 - JUST ! (without any patches)) with tcpwrapper ucspi-0.88 and multilog from daemontools package (v. 0.70) all them compiled with gcc 2.95.2 running in 4 machines PIII - 500Mhz with 256mb RAM. In my DNS I have smtp.ig.com.br pointing to IPs addresses of that four machines trying to make some kind of load balance (Probably soon I will replace this by IPVS or something like that (Altheon, etc). The total amount of messages that are sent through those machines are aprox. 1.5 million by day (!) . I have set remote and local concurrency to 255 (and I change the limit in conf-spawn to 255 and recompiled qmail after that). My tcpwrapper is configured to 400 simultaneous smtp connections (per machine). The remote and localconcurrency stay between 30-150 connections so, I think that 255 limit isn't problem by now (Maybe in future I will need to install the patch to concurrency above 400). Memory and wait I/O isn't the problem. The number of file descriptors are ok (2048) now (Now I don't receive messages tell me that was not possible to open a pipe (or something like that)). The problem is the fact that I'm having too much defunct process. Usually I have between 350-500 process running by machine. From that, normally I have between 90-120 defunct process per machine. Anybody have ANY idea about what can be causing this ? Is there any patch to correct this problem, in case of this be a well know problem for my plataform ? PS. multilog and tcpwrapper are in PATH and this problem NOT started after I reboot my machines ;-) Any help will be very appreciated . Thanks in advance... Claudio Cuqui [EMAIL PROTECTED] System Analyst Internet Group do Brasil http://www.ig.com.br
Re: Problems using qmail on very large site
The problem is the fact that I'm having too much defunct process. Usually I have between 350-500 process running by machine. From that, normally I have between 90-120 defunct process per machine. Do they stay forever, or do they go away? Which qmail process is the parent? I have seen Solaris 2.x systems where processes do stay around forever, but I have not seen in on, eg, FreeBSD. Regards.
