Re: [SSI-users] Cluster strange boot behaviour
Brought to you by:
brucewalker,
rogertsang
From: Brian J. W. <Bri...@hp...> - 2005-02-16 23:44:57
|
S=E9bastien Gross wrote: > Now after the cluster booted up, I launched serveral applications to > test the stability. On each node (smp servers): > - load: 6~10 > - cpu: user 30%, system: 30%, nice 140%, idle 0%. > - memory 1~2 GB on a 4GB memory set > - no swap used. >=20 > after a couple of time (between 2h to a night) the whole cluster crashe= s > down. > In fact running processes a still usable (such as apache) but no other > new process can be launched from anywhere (even console login is not > launched). >=20 > Does any one experienced such very abnormal behaviour? It's hard to guess what might be going wrong. The next time you=20 experience this, a netdump image from one or more of the nodes might=20 help debug this. > One other question about services such as apache: I put /usr/sbin/apach= e > in the loadlevellist configuration, works great. But process seem to be > migrated _before_ the port 80 is opened on the init node, so as all > nodes are behind (masqueraded) by node 1, apache is listening on node 4 > for example. > Is there a way to force port opening on init (or gateway) node(s) ? Configure your cluster to manually start an instance of apache on every=20 node. Then use TCP connection load balancing (LVS) to load-level=20 connection requests among the instances of apache. I'm certain this has=20 been discussed before on this list (or ssic-linux-devel), so more=20 information can be found in the archives. Regards, Brian |