|
btw in our case we fetch a static object which is much faster.
Sameer
On Thu, Mar 5, 2009 at 6:37 PM, Sameer Garg <sameer.garg@xxxxxxxxx> wrote:
> Hey Brandon,
>
> The slow startup problem then lies with the real server. When
> ldirectord is started it does a health check and adds machines to LVS.
> So depending upon the page loading time your LVS startup time will
> vary.
>
> Sameer
>
> On Sat, Feb 7, 2009 at 8:50 PM, Brandon Hilkert <bhilkert@xxxxxx> wrote:
>> Hey Sameer,
>>
>> We're just using negotiate, not both. Sorry if I made it confusing. Part of
>> the problem is that most of our websites are database driven and then page
>> we check loads a bunch of data objects, so sometimes during high traffic
>> load, it may take 15-20 sec for the page to load, so I have to increase the
>> negotiate timeout to something much higher to account for this, or else the
>> thing would be failing over constantly. For us, a simple connect check
>> wouldn't necessarily be a true test if the website is functioning properly.
>> The IP could be up and pass the Ldirector test, but if the page doesn't
>> display properly, then it's useless and should certainly failover. We're
>> balancing windows machines, and sometimes and they don't just fail like
>> that, the IP will be up, but the website wouldn't be displaying.
>> Unfortunate, but true...
>>
>> I've tried to pair the negotiate timeouts with the site itself on a
>> case-by-case basis hoping to minimize the delay during startup and failover.
>> It seems I don't have much choice. I guess I just have to hope that the
>> Ldirectord doesn't randomly go down, but that's why it's on Linux right? :)
>>
>> Thanks,
>> Brandon
>> ----- Original Message -----
>> From: "Sameer Garg" <sameer.garg@xxxxxxxxx>
>> To: "LinuxVirtualServer.org users mailing list."
>> <lvs-users@xxxxxxxxxxxxxxxxxxxxxx>
>> Sent: Friday, February 06, 2009 11:06 PM
>> Subject: Re: [lvs-users] Ldirector Startup Delay
>>
>>
>> Hi Brandon,
>>
>> I am using LVS in a scenario where we are load balancing 20 different
>> subdomains across 40+ machines and have yet to experience a delay. Why
>> are you using two types of checks: negotiate and http check? I use
>> http and it works like a charm.
>>
>> Sameer
>>
>> On Wed, Jan 14, 2009 at 1:44 AM, Brandon Hilkert <bhilkert@xxxxxx> wrote:
>>> We're trying to balance 12 or so sites off this one ldirector box. Because
>>> we use the negotiate check method and check for a keyword on a page,
>>> sometimes there is a few minute delay when the service starts because it
>>> goes site by site and checks each one, then brings it up upon successfully
>>> find the keyword.
>>>
>>> Is there any way to have the service automatically bring everything up and
>>> then go into its checking routine, but first assume all the real servers
>>> are up?
>>>
>>> This would help us avoid a few minutes of downtime if the machine/service
>>> were to be restarted.
>>>
>>> Thanks,
>>> Brandon
>>> _______________________________________________
>>> Please read the documentation before posting - it's available at:
>>> http://www.linuxvirtualserver.org/
>>>
>>> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>>> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>>> or go to http://lists.graemef.net/mailman/listinfo/lvs-users
>>>
>>
>>
>>
>>
>> _______________________________________________
>> Please read the documentation before posting - it's available at:
>> http://www.linuxvirtualserver.org/
>>
>> LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
>> Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
>> or go to http://lists.graemef.net/mailman/listinfo/lvs-users
>>
>
_______________________________________________
Please read the documentation before posting - it's available at:
http://www.linuxvirtualserver.org/
LinuxVirtualServer.org mailing list - lvs-users@xxxxxxxxxxxxxxxxxxxxxx
Send requests to lvs-users-request@xxxxxxxxxxxxxxxxxxxxxx
or go to http://lists.graemef.net/mailman/listinfo/lvs-users
|
