Hi Simon, On Tue, 2010-12-21 at 07:41 +0900, Simon Horman wrote:
> while there are a number of limitations in the synchronisation code > I believe that what you are trying to do should work. On the backup > do you see connections showing up in the output of ipvsadm -Lcn ? They don't, in the problematic cases. They do for the direct (non-nat / non-fwmark) case, i.e. the setup is working in principle. I decorated ip_vs_sync:ip_vs_sync_conn with a debug_level 7 output at the end. The normal, working and syncing case shows: Dec 21 08:49:43 gw1 kernel: [71248.885908] IPVS: ip_vs_sync_conn with len 24, curr_sb ffff880076c586c0 Dec 21 08:49:43 gw1 kernel: [71248.953799] IPVS: TCP input [.FA.] 10.0.12.33:80->X.X.X.X:45986 state: FIN_WAIT->TIME_WAIT conn->refcnt:2 Dec 21 08:49:43 gw1 kernel: [71248.953848] IPVS: ip_vs_sync_conn with len 24, curr_sb ffff880076c586c0 Dec 21 08:49:43 gw1 kernel: [71249.290038] IPVS: Enter: ip_vs_send_async, net/netfilter/ipvs/ip_vs_sync.c line 707 Dec 21 08:49:43 gw1 kernel: [71249.290080] IPVS: Leave: ip_vs_send_async, net/netfilter/ipvs/ip_vs_sync.c line 713 The connection appears on the backup server (*) For the case with DNAT I only always see this: Dec 21 08:51:04 gw1 kernel: [71329.419629] IPVS: TCP output [..A.] 10.0.12.33:80->X.X.X.X:48911 state: SYN_RECV->ESTABLISHED conn->refcnt:2 Dec 21 08:51:09 gw1 kernel: [71334.789943] IPVS: TCP output [.FA.] 10.0.12.33:80->X.X.X.X:48911 state: ESTABLISHED->FIN_WAIT conn->refcnt:2 and nothing turns up. However, I just found that for the fwmark case, today, I do see the send_async debug output, AND I now do see the connections on the backup server. I'm 80% sure that was different yesterday when I sent the first mail, but maybe I confused myself... The DNAT case is not syncing in any case. I will not try to go with the fwmark approach and see whether it stays stable. > On the topic of connection synchronisation, a new synchronisation > protocol (v1) has been developed which addresses a number of the problems > in the existing code (v0). This includes the explicit synchronisation > of fwmarks. The code is currently available in the lvs-test-2.6 tree > on git.kernel.org. It should appear in 2.6.38-rc1. Thanks for the pointer. Unfortunately I'm not at all familiar with git and using such development trees, right now, and I'm a bit reluctant to try something like that on a soon-to-be-production box. So I'll look forward to testing that in spring when 2.6.38 is a bit settled. _______________________________________________ Please read the documentation before posting - it's available at: http://www.linuxvirtualserver.org/ LinuxVirtualServer.org mailing list - lvs-users@LinuxVirtualServer.org Send requests to lvs-users-requ...@linuxvirtualserver.org or go to http://lists.graemef.net/mailman/listinfo/lvs-users
