eviden-logo

Evidian > Products > High Availability Software - Zero Extra Hardware > Heartbeat, failover and quorum in a Windows or Linux cluster

Heartbeat, failover and quorum in a Windows or Linux cluster

Evidian SafeKit

What are the different scenarios in case of network isolation in a cluster?

A single network

When there is a network isolation, the default behavior is:

  • as heartbeats are lost for each node, each node goes to ALONE and runs the application with its virtual IP address (double execution of the application modifying its local data),
  • when the isolation is repaired, one ALONE node is forced to stop and to resynchronize its data from the other node,
  • at the end the cluster is PRIM-SECOND (or SECOND-PRIM according the duplicate virtual IP address detection made by Windows).

Two networks with a dedicated replication network

When there is a network isolation, the behavior with a dedicated replication network is:

  • a dedicated replication network is implemented on a private network,
  • heartbeats on the production network are lost (isolated network),
  • heartbeats on the replication network are working (not isolated network),
  • the cluster stays in PRIM/SECOND state.

A single network and a splitbrain checker

When there is a network isolation, the behavior with a split-brain checker is:

  • a split-brain checker has been configured with the IP address of a witness (typically a router),
  • the split-brain checker operates when a server goes from PRIM to ALONE or from SECOND to ALONE,
  • in case of network isolation, before going to ALONE, both nodes test the IP address,
  • the node which can access the IP address goes to ALONE, the other one goes to WAIT,
  • when the isolation is repaired, the WAIT node resynchronizes its data and becomes SECOND.

Note: If the witness is down or disconnected, both nodes go to WAIT and the application is no more running. That's why you must choose a robust witness like a router.

How heartbeats and failover work in a Windows or Linux cluster?

What is a heartbeat?

The basic mechanism for synchronizing two servers and detecting server failures is the heartbeat, which is a monitoring data flow on a network shared by a pair of servers.

The SafeKit software supports as many heartbeats as there are networks shared by two servers. 

The heartbeat mechanism is used to implement Windows and Linux clusters. It is integrated within the SafeKit mirror cluster with real-time file replication and failover.

SafeKit heartbeats

In normal operation, the two servers exchange their states (PRIM, SECOND, the resource states) through the heartbeat channels and synchronize their application start and stop procedures.

In particular, in case of a scheduled failover, the stop script which stops the application is first executed on the primary server, before executing the start script on the secondary server. Thus, replicated data on the secondary server are in a safe state corresponding to a clean stop of the application.

Loss of all heartbeats

When all heartbeats are lost on one server, this server considers the other server to be down and transitions to the ALONE state.

If it is the SECOND server which goes to the ALONE state, then there is an application failover with restart of the application on the secondary server.

Although not mandatory, it is better to have two heartbeat channels on two different networks for synchronizing the two servers in order to separate the network failure case from the server failure one.

Split brain problem and quorum when servers are in two remote computer rooms

Heartbeat, failover and quorum in a Windows or Linux cluster

Remote computer rooms

A high availability cluster securing a critical application can be implemented with two servers in two geographically remote computer rooms.

Thus, the solution supports the disaster of a full room.

Split brain

In situation of a network isolation between both computer rooms, all heartbeats are lost and the split brain problem arises.

Both servers start the critical application.

Complexity of solutions

Mostoften, to solve split brain, quorum is implemented with a third quorum server or a special quorum disk to avoid the double masters.

Unfortunately these new quorum devices add cost and complexity to the overall clustering architecture.

Simple cluster quorum with the SafeKit split brain checker

SafeKit split brain checker

With the SafeKit high availability software, the quorum within a Windows or Linux cluster requires no third quorum server and no quorum disk. A simple split brain checker is sufficient to avoid the double execution of an application.

On the the loss of all heartbeats between servers, the split brain checker selects only one server to become the primary. The other server goes into the WAIT state, until it receives the other server's heartbeats again. It then goes back to secondary after having synchronized replicated data from the primary server.

How the split brain checker works?

The primary server election is based on the ping of an IP address, called the witness. The witness is typically a router that is always available. In case of network isolation, only the server with access to the witness will be primary ALONE, the other will go to WAIT.

The witness is not tested permanently but only when all heartbeats are lost. If at that time, the witness is down, the cluster goes into the WAIT-WAIT state and an administrator can choose to restart one of the servers as primary through the SafeKit web console.

What happens without a split brain checker?

In case of network isolation, both servers will go to the ALONE state running the critical application. The replicated directories are isolated and each application is working on its own data in its own directory.

When the network is reconnected, SafeKit by default chooses the server which was PRIM before the isolation as the new primay and forces the other one as SECOND with a resynchronization of all its data from the PRIM.

Note: Windows can detect a duplicate IP address on one server and remove the virtual IP address on this server. SafeKit has a checker to force a restart in that case.

How the SafeKit mirror cluster works?

Step 1. Real-time replication

Server 1 (PRIM) runs the application. Clients are connected to a virtual IP address. SafeKit replicates in real time modifications made inside files through the network. 

File replication at byte level in a mirror cluster

The replication is synchronous with no data loss on failure contrary to asynchronous replication.

You just have to configure the names of directories to replicate in SafeKit. There are no pre-requisites on disk organization. Directories may be located in the system disk.

Step 2. Automatic failover

When Server 1 fails, Server 2 takes over. SafeKit switches the virtual IP address and restarts the application automatically on Server 2.

The application finds the files replicated by SafeKit uptodate on Server 2. The application continues to run on Server 2 by locally modifying its files that are no longer replicated to Server 1.

Failover in a mirror cluster

The failover time is equal to the fault-detection time (30 seconds by default) plus the application start-up time.

Step 3. Automatic failback

Failback involves restarting Server 1 after fixing the problem that caused it to fail.

SafeKit automatically resynchronizes the files, updating only the files modified on Server 2 while Server 1 was halted.

Failback in a mirror cluster

Failback takes place without disturbing the application, which can continue running on Server 2.

Step 4. Back to normal

After reintegration, the files are once again in mirror mode, as in step 1. The system is back in high-availability mode, with the application running on Server 2 and SafeKit replicating file updates to Server 1.

Return to normal operation in a mirror cluster

If the administrator wishes the application to run on Server 1, he/she can execute a "swap" command either manually at an appropriate time, or automatically through configuration.

Typical usage with SafeKit

Why a replication of a few Tera-bytes?

Resynchronization time after a failure (step 3)

  • 1 Gb/s network ≈ 3 Hours for 1 Tera-bytes.
  • 10 Gb/s network ≈ 1 Hour for 1 Tera-bytes or less depending on disk write performances.

Alternative

Why a replication < 1,000,000 files?

  • Resynchronization time performance after a failure (step 3).
  • Time to check each file between both nodes.

Alternative

  • Put the many files to replicate in a virtual hard disk / virtual machine.
  • Only the files representing the virtual hard disk / virtual machine will be replicated and resynchronized in this case.

Why a failover ≤ 32 replicated VMs?

  • Each VM runs in an independent mirror module.
  • Maximum of 32 mirror modules running on the same cluster.

Alternative

  • Use an external shared storage and another VM clustering solution.
  • More expensive, more complex.

Why a LAN/VLAN network between remote sites?

Alternative

  • Use a load balancer for the virtual IP address if the 2 nodes are in 2 subnets (supported by SafeKit, especially in the cloud).
  • Use backup solutions with asynchronous replication for high latency network.

SafeKit High Availability Differentiators

SafeKit Solutions Overview

SafeKit Quick Installation Guides

New application (real-time replication and failover)

New application (network load balancing and failover)

Database (real-time replication and failover)

Web (network load balancing and failover)

Full VM or container real-time replication and failover

Amazon AWS

Google GCP

Microsoft Azure

Other clouds

Physical security (real-time replication and failover)

Siemens (real-time replication and failover)