cloudstack-dev mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From Chiradeep Vittal <Chiradeep.Vit...@citrix.com>
Subject Re: [DISCUSS] Method/Algorithm to gather Health Check states
Date Fri, 08 Mar 2013 00:40:22 GMT
I don't see this requirement in the original requirement
https://cwiki.apache.org/confluence/display/CLOUDSTACK/Load+Balancer+Health
+checks


IMO, this should be a different service (monitoring service) that can be
report on LB status.

I.E, this is not part of this feature.

On 3/7/13 7:51 AM, "Vijay Venkatachalam" <Vijay.Venkatachalam@citrix.com>
wrote:

>
>I have mentioned Pros and Cons, thanks for asking!.
>
>Hopefully, it gives a better picture now
>
>> -----Original Message-----
>> From: Chip Childers [mailto:chip.childers@sungard.com]
>> Sent: Thursday, March 07, 2013 7:40 PM
>> To: cloudstack-dev@incubator.apache.org
>> Subject: Re: [DISCUSS] Method/Algorithm to gather Health Check states
>> 
>> On Thu, Mar 07, 2013 at 07:17:17PM +0530, Vijay Venkatachalam wrote:
>> >
>> > Any  votes/re-assurances to the thought process from the Networking
>> Universe?
>> >
>> > My Vote: Method-2
>> >
>> > -Vijay
>> 
>> So I'm struggling to understand the implications of doing it via method
>> 1 vs method 2.  Can you provide what you think the pros/cons are?
>> 
>> >
>> > -----Original Message-----
>> > From: Vijay Venkatachalam
>> > Sent: Wednesday, March 06, 2013 7:58 PM
>> > To: cloudstack-dev@incubator.apache.org
>> > Subject: [DISCUSS] Method/Algorithm to gather Health Check states
>> >
>> > Hi,
>> >
>> > As part of HealthCheck feature review, I have suggested to introduce a
>> new Capability for HealthCheck. So that any Network Element who has this
>> capability can advertise it.
>> >
>> > Any alternative thoughts?
>> >
>> > The most important part of health check functionality is to update the
>> status of a destination VM in CS DB (as seen by the LB Appliance) on a
>> scheduled thread launched every time interval.
>> >
>> > So on every iteration, the need is to go to every LB rule and find
>>the status
>> of its destinations.
>> >
>> > I can think of 2 ways to get this done
>> >
>> > Method 1:
>> >   A.  List all the Health Monitors; For each monitor=>
>> >   B.  Work backwards find the LbRule
>> >   C. Find the network for LbRule
>> >   D. Find the LB provider in the network
>> >   E. Call LoadBalancingServiceProvider.updateStatus by passing LB and
>>its
>> destinations.
>> >   F. Which will reach the Resource and ultimately the Appliance
>> >
>
>Pros:
>1.  Right in the beginning you have the LbRules in the entire system, so
>you 
>     have a perfect working set for which status have to be collected.
>
>Cons:
>1.  Scope of narrowing down logically to work on the problem set is not
>present.
>
>> > Method 2:
>> >   A. List all the Networks; For each network =>
>> >   B. Find the LB provider in the network
>> >   C. Find the NetworkElement for the LB provider
>> >   D. Proceed to (E.)  If NetworkElement has the HealthCheck Capability
>> >   E. Call LoadBalancingServiceProvider.updateStatus by passing LB and
>>its
>> destinations.
>> >   F. Which will reach the Resource and ultimately the Appliance
>> >
>
>Pros:
>1.  Logical break down of the working set "Network" -> "LBRules" ->
>"Appliance". 
>2.  Possibility of optimization because multiple LB rules of the network
>can be 
>    passed together to Network Element and NetworkElement can optimize
>    by sending all lbrules destined to an Appliance together to the
>Resource for status.
>3.  Also, in case if we decide to report the status of destination VMs
>even 
>    if HealthMonitors are not specifically configured by the user (this
>can be done 
>    because almost all LB appliances track the status of destinations by
>binding 
>    some default monitor to check connectivity at least to the port),
>this is the only way.
>4. Is there a division of responsibility when multiple ManagementServers
>are installed, 
>     Is it based on networks? Then this approach makes more sense
>
>Cons:
>1.  Even if no LB rules are there with health monitors we will be doing
>Steps A-D
>
>> > I like the top down approach of Method 2. Any thoughts?
>> >
>> > Thanks,
>> > Vijay V.
>> >
>> >
>> > From: Vijay Venkatachalam [mailto:noreply@reviews.apache.org] On
>> Behalf Of Vijay Venkatachalam
>> > Sent: Wednesday, March 06, 2013 6:10 PM
>> > To: Vijay Venkatachalam
>> > Cc: Rajesh Battala; cloudstack
>> > Subject: Re: Review Request: AWS Style HealthCheck feature BugID : 664


Mime
View raw message