"Cloud" do it yourself

mayorovp2012-01-25 21:22:32

linux

mayorovp, 2012-01-25 21:22:32

The task is as follows: there is a set of services and two physical servers.
How it is possible to organize automatic start of all services on the second server in case of failure of the first?

I'm not interested in "bicycles" on bash, but in standard solutions.
Who knows the answer to this question, please help, otherwise I have been trying for two hours to find out if Xen Server and VmWare Server support this, but I have not found a clear answer. If anyone knows solutions without using virtualization, I'll be glad too.

Answer the question

In order to leave comments, you need to log in

9 answer(s)

alexk24, 2012-01-26
@mayorovp

Do you need a transition to a reserve after a service or server crash?
If the servers are XCP or vSphere.
If services - you will have to make a failover cluster in accordance with the recommendations for this service. Because neither XCP nor vSphere is designed to control individual services. But this is if you are lucky and clustering this service by standard means is possible. Otherwise, just "bike on bash".
In general, the task of reserving a service is most likely not solved. Need to know the details.

cat_crash, 2012-01-25
@cat_crash

In pursuit - what does your question have to do with clouds?

cat_crash, 2012-01-25
@cat_crash

Do you need only services or services + data? This is an important question, because it’s not a problem to start services through watchdog, but it’s more difficult to synchronize and update data

Sergey, 2012-01-26
@bondbig

As already mentioned, the task never describes a “cloud” (what a mess in people’s heads, horror), but quite a boring, long-standing HA cluster. The implementation will greatly depend on what exactly is planned to be played on the servers.

mikes, 2012-01-25
@mikes

Vmware server has not been supported for a long time ... the functionality you need is in vmware vshere and in several options
1 option - the machine migrates to another physical server in case of failure (requires a separate datastore and for clients the failure looks like a sudden server reboot)
Option 2 - continuous replication ... machine launched in 2 instances and the RAM is synchronized ... if it fails, there are no losses, respectively.

lexeresser, 2012-01-26
@lexeresser

drdb?

Puma Thailand, 2012-01-26
@opium

Use the standard live migration to xen or vmware.

cepera_ang, 2012-07-12
@cepera_ang

Indeed, a banal HA cluster is required. Vmware still calls it HA-cluster =). It is done very simply - there is a storage on which virtual machines lie, there are two or more servers, they are added to the cluster, all the necessary checkboxes are put, virtual machines are started, everything works. If one of the servers crashes for some reason, all VMs on the second one are restarted from the first one. For users, it looks like the VMs just rebooted.
If the services you use themselves support some kind of clustering, then you can run the corresponding halves of the cluster on different hosts + enable HA. Then, when the server crashes, both parts of the cluster will start on one (although this will no longer be fault-tolerant), when the service crashes, its own cluster will work.