|
Hi Ratz,
On Tue, 1 Jul 2003, Roberto Nibali wrote:
>
> Hey, if you're already in a mood of accepting so many changes what do
> you think if we just include the whole per RS threshold limitation
> backport to 2.4.x once it's tested enough?
>
If the backport to ipvs 1.0.x for 2.4.x is very well tested, we will
include it in the next release of ipvs 1.0.x. We usually don't change ipvs
1.0.x too much, because 1.0.x is the stable branch.
> For that one I also have a question: You probably remember that I used
> to quiesce all RS when upper threshold was reached in 2.2.x days. My
> user space app could then determine with a simple awk if all RS of a
> service template are quiesced and thus if the whole service would be
> overloaded. It would then invoke a session overload page on localhost
> via local route forward. And if the RS service was not reachable anymore
> through one of the various healthchecks the user space app would take it
> out and in case of a complete outtake of all RS on a service template
> invoke a page of last resort indicating that the service was currently down.
>
> The page of last resort takes precedence over the session overload page
> unless persistency is enabled for the latter (obviously). What's
> important to realise is that I did a distinct separation between a
> session overload and a service takeout!
>
> Now with the new implementation of the per RS threshold limitation I do
> not get any indication in user space which shows me the status, as
> internally only the IP_VS_DEST_F_OVERLOAD flag is set and if it is set,
> the scheduler doesn't pick this dest. Which is basically the same, as
> seen from user space, like taking the service out with ipvsadm -d ... .
>
> How do you think could I solve this problem of mine?
>
Maybe there are two solutions to it.
1) add the flags member into the struct ip_vs_dest_entry, so that the
status flags of real server can be exported to user-space programs.
2) add the failover server into each virtual service entry. When there
is no server available, send the request to the failover server.
The failover server can be run at the director machine, or the
remote machine.
I am going to implement the latter in ipvs 1.1.x.
Cheers,
Wensong
|
