• No se han encontrado resultados

CLINICAL AND PROGNOSTIC PROFILE OF PATIENTS WITH INFECTIVE ENDOCARDITIS WHO NEED URGENT SURGERY.

6 RESUMEN DE LOS TRABAJOS

CLINICAL AND PROGNOSTIC PROFILE OF PATIENTS WITH INFECTIVE ENDOCARDITIS WHO NEED URGENT SURGERY.

Although some IBM middleware such as DB2 (and InfoSphere Warehouse) come with HA capability, many of the software services listed in this section do not have native HA features. Therefore, these features might have to be sourced from third-party software. Although virtualization and cloud computing resources provide HA from a server-level standpoint, HA can also be provided at the

service

level.

The advantages of additionally providing HA at the service level are quicker recovery time, more fault tolerant services, and reducing strain on the virtualized environment. Server resources at the virtualization level do not need to be

redeployed because the underlying services were “healed.” Consider software that is capable of providing both HA and automated failover capabilities.

High availability and automated failover for services

Software, such as Tivoli System Automation for Multiplatforms and LinuxHA, can provide HA on just about any service or daemon running within the network. Within any operating system are services and daemons that are running. Automation software can query these services to check whether the status of each service is running, stopped, or if there are problems. LDAP, NTP, DNS, web server, database, and Cognos are all examples of services. You can configure this automation software to perform a variety of operations if a failure occurs on the services that its designed to maintain.

Consider the following possible configurations for HA: 򐂰 Active/cold standby

– One server handles active connections. The other server is “sleeping” as standby and is ready to take over as the active server in a failover scenario.

– This configuration provides the lowest manageability requirements and the lowest resource efficiency.

򐂰 Active/hot standby (master/slave)

– One server handles active connections (master). The other server has a slave process of its paired active server.

– This configuration provides moderate manageability requirements and low resource efficiency.

򐂰 Active/active (master/master)

– An active/active configuration load balances servers within the

configuration. When a server fails, the remaining healthy server or servers continue processing connections. This HA configuration usually requires a homogenous software configuration and higher manageability

requirements, but also yields the highest resource efficiency because all servers are being used. This configuration provides the highest

manageability requirements and the best resource efficiency.

From these HA configuration choices, determine the most appropriate method for deploying your HA clustered software components. Automation software has many configurable parameters for fine-tuning HA and failover. By implementing a sound design for your HA software, you can ensure that your solution’s

underlying services are running healthy, thereby allowing your solution to serve its customers unhindered.

Distributed Replicated Block Device (DRBD) is a distributed storage system for the Linux platform, as shown in Figure 3-16. It provides HA for storage (block devices) in either master/slave or master/master configurations. DRBD involves replication between two or more nodes.

DRBD is ideal for integrating shared storage file system in software services that use master/slave or active/passive HA configuration such as DNS and LDAP. This is due to the fact that both master and slave can access the same file system in the event of a failure. DRBD has mechanisms to ensure that the data stored on the DRBD device is consistent and accurate.

Figure 3-16 DRBD implementation

Example high availability components implementation

Consider the HA solution depicted in Figure 3-17 on page 138. This solution has the following configuration:

򐂰 Services

– LDAP, DNS, HA Proxy (web server load balancer), and Clustered (virtual) IP services are handled by a generic automation software and DRBD configuration. Whenever a service is detected to be not running or unhealthy, the service on the standby server is brought up to resume work.

– Services and the Cognos gateway reside in the same physical machine as an active/standby pair. The active server contains all active services and

DRBD/Primary DRBD/Secondary hda1 hda2 nb0 hda1 hda2 hdb1 DRBD hdb1 hda1 hda2 Nb0 hda1 hda2 hdb1 hdb1 Nb0 R/W Access R/W Access Synchronized DRBD DRBD/Primary DRBD/Secondary

active Cognos gateway. The standby server contains the same services as the active server, but on standby, though Cognos Gateway on this server is also active (so that the web server load balancer can feed incoming clients to either the active or standby machines).

Figure 3-17 High availability implementation

– Services directories and configuration files can change due to ongoing configuration and addition of clients and servers. These files and directories are stored on a DRBD device that can be read by only the current active server. Setup of the DRBD device is as follows:

• Create partitions on both primary and standby.

• Create logical volumes (size determined by service space requirements) on both primary and standby.

• Create an DRBD device that encompass the logical volumes on both primary and standby.

• Determine which service files and directories must be placed on the DRBD device and place them there.

Failure scenarios

These HA configurations can help mitigate downtime of an overall solution by providing automated failover capabilities. The automated failover capabilities kick in when a service or hardware component malfunctions. The following sample scenarios show how the HA software reacts when certain failures occur: 򐂰 Failure Scenario 1: Service failure

If a particular service is stopped unexpectedly and cannot start again, the automation software fails over to the standby server. The DRBD device is made readable on the new active server.

All services are started on the standby server, and the cluster IP now points to the new active server. All requests are redirected to the standby server. HA Proxy can still direct Cognos traffic to the old active server because the private network is still available. The Cognos gateway is still running. 򐂰 Failure Scenario 2: Cognos Gateway failure

The automation software does not do anything. The only change is that HA Proxy stops directing Cognos traffic to the standby server’s Cognos gateway. 򐂰 Failure Scenario 3: Network failure on active server

A split-brain scenario is possible if there is no network quorum.

Split-brain

occurs when communication between the primary and standby is terminated, making both the primary and the standby servers think each server should take over operations.

To avoid split-brain scenarios, a network quorum must be achievable on both servers. A network quorum should be an address in the network that both servers can ping and that is reliable. If the automation software on a particular server cannot achieve quorum, all services are shut down. The standby server still has quorum, can detect that the active server is unreachable, and can take over accordingly.

򐂰 Failure Scenario 4: Power failure on active server

The automation software detects a time out on the active server and makes the standby the new active server. The DRBD device and all services are made active on the new active server.