QUORUM-BASED POWER-DOWN OF UNRESPONSIVE SERVERS IN A COMPUTER CLUSTER

• US 20080301272A1
• Filed: 08/15/2008
• Published: 12/04/2008
• Est. Priority Date: 11/04/2004
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Jones, Christopher Henry, Wallis, Graham Derek, Newport, William T.
• Granted Patent: US 7,716,222 B2
• Time in Patent Office: Days
• US Class Current: 709/221
• CPC Class Codes: H04L 12/12 Arrangements for remote connection or disconnection of substations or of equipment thereof
• CPC Class Codes: H04L 41/0681 Configuration of triggering conditions
• CPC Class Codes: H04L 43/10 Active monitoring, e.g. heartbeat, ping or trace-route
• CPC Class Codes: Y02D 30/50 in wire-line communication networks, e.g. low power modes or reduced link rate
• CPC Class Codes: Y10S 707/955 Object-oriented

Abstract

A quorum-based server power-down mechanism allows a manager in a computer cluster to power-down unresponsive servers in a manner that assures that an unresponsive server does not become responsive again. In order for a manager in a cluster to power down servers in the cluster, the cluster must have quorum, meaning that a majority of the computers in the cluster must be responsive. If the cluster has quorum, and if the manager server did not fail, the manager causes the failed server(s) to be powered down. If the manager server did fail, the new manager causes all unresponsive servers in the cluster to be powered down. If the power-down is successful, the resources on the failed server(s) may be failed over to other servers in the cluster that were not powered down. If the power-down is not successful, the cluster is disabled.

Claims

1. A computer-implemented method for handling an unresponsive server in a cluster, the method comprising the steps of:

determining when a server in the cluster becomes unresponsive;

determining whether a majority of servers in the cluster are responsive;

receiving an indication of a server failure;

determining whether a manager of the cluster failed;

if a manager of the cluster failed when the indication of a server failure is received, issuing at least one command to power down all unresponsive servers in the cluster, wherein an unresponsive server is a server that fails to send a periodic message that indicates the server is functioning properly; and

if a manager of the cluster did not fail when the indication of a server failure is received, issuing at least one command to power down a server corresponding to the received indication of server failure.

2. The method of claim 1 further comprising the step of each server in the cluster sending a periodic message to other servers in the cluster to indicate proper function of the server sending the periodic message.

3. The method of claim 1 further comprising the step of each server receiving periodic messages from other servers in the cluster that indicate the other servers in the cluster are functioning properly.

4. The method of claim 3 further comprising the step of generating a membership change message to all servers in the cluster when a selected server in the cluster becomes unresponsive by failing to send a periodic message that indicates the selected server is functioning properly.

5. The method of claim 1 further comprising the step of generating a membership change message to all servers in the cluster when any of the servers in the cluster become unresponsive.

6. The method of claim 1 further comprising the steps of:

determining whether the power down of the at least one of the other servers was successful;

if the power down of the at least one of the other servers was successful, enabling the failing over any resources on the at least one of the other servers that was powered down to at least one server that is responsive; and

if the power down of the at least one of the other servers was not successful, disabling the cluster.

7. A computer-implemented method for handling an unresponsive server in a cluster, the method comprising the steps of:

each server in the cluster sending a periodic message to other servers in the cluster to indicate proper function of the server sending the periodic message;

each server in the cluster receiving periodic messages from other servers in the cluster that indicate the other servers in the cluster are functioning properly;

generating a membership change message to all servers in the cluster when any of the servers in the cluster become unresponsive;

determining whether a majority of servers in the cluster are responsive;

receiving an indication of a server failure;

if the majority of servers in the cluster are responsive, performing the steps of;

determining whether the indication of the server failure indicates a manager of the cluster failed;

if the manager of the cluster failed, issuing at least one command to power down all unresponsive servers in the cluster, wherein an unresponsive server is a server that fails to send a periodic message that indicates the server is functioning properly; and

if the manager of the cluster did not fail, issuing at least one command to power down a server corresponding to the received indication of server failure.

8. The method of claim 7 further comprising the steps of:

determining whether the power down of the at least one of the other servers was successful;

if the power down of the at least one of the other servers was successful, enabling the failing over any resources on the at least one of the other servers that was powered down to at least one server that is responsive; and

if the power down of the at least one of the other servers was not successful, disabling the cluster.