Linux-HA Discharge 2 Instructional exercise.

Uploaded on:
Category: Education / Career
Fencing tries to put a wall around an errant hub or hubs to keep them from getting to bunch assets ... Else, we just fence hubs that leave the bunch subsequent to having ...
Slide 1

Linux-HA Release 2 Tutorial Alan Robertson Project Leader – Linux-HA venture IBM Systems & Technology Group Industry Technology Leadership Team HA Subject Matter Expert

Slide 2

Tutorial Overview HA Principles Installing Linux-HA Basic Linux-HA arrangement Configuring Linux-HA Sample HA Configurations Testing Clusters Advanced components

Slide 3

Part I General HA standards Architectural outline of Linux-HA Compilation and establishment of the Linux-HA ("heartbeat") programming

Slide 4

What Is HA Clustering? Assembling a gathering of PCs which believe each other to give an administration notwithstanding when framework parts fizzle When one machine goes down, others assume control over its work This includes IP address takeover, administration takeover, and so on. New work goes to the "takeover" machine Not essentially intended for superior

Slide 5

What Can HA Clustering Do For You? It can\'t accomplish 100% accessibility – nothing can. HA Clustering intended to recoup from single issues It can make your blackouts short From around a second to a couple of minutes It resemble a Magician\'s (Illusionist\'s) trap: When it goes well, the hand is quicker than the eye When it goes not really well, it can be sensibly noticeable A decent HA grouping framework includes a "9" to your base accessibility 99->99.9, 99.9->99.99, 99.99->99.999, and so forth. Intricacy is the foe of unwavering quality!

Slide 6

High-Availability Workload Failover

Slide 7

Lies, Damn Lies, and Statistics Counting nines

Slide 8

How is this like what you know? It\'s a ton like the current init startup scripts reached out by (alternatively) including parameters to them running a more than one PC adding approaches for what request to do things how benefits identify with each other when to run them HA frameworks are a considerable measure like "init on steroids"

Slide 9

What is distinctive? HA Clusters present ideas and entanglements around Split-Brain Quorum Fencing Data offering isn\'t normally an issue to a solitary server – it\'s basically vital in groups

Slide 10

Split-Brain Communications disappointments can prompt isolated parcels of the bunch If those segments every attempt and take control of the bunch, then it\'s known as a split-mind condition If this happens, then awful things will happen

Slide 11

Quorum is an endeavor to maintain a strategic distance from split cerebrum for most sorts of disappointments Typically one tries to ensure one and only segment can be dynamic Quorum is term for strategies for guaranteeing this Most regular sort of majority is voting – and just a segment with > n/2 hubs can run the bunch This doesn\'t work exceptionally well for 2 hubs :- (

Slide 12

Fencing tries to put a wall around an errant hub or hubs to keep them from getting to group assets This way one doesn\'t need to depend on right conduct or timing of the errant hub. We utilize STONITH to do this STONITH: Shoot The Other Node In The Head Other systems additionally work Fiber channel switch lockout and so forth

Slide 13

How is HA Clustering Different from Disaster Recovery ("geographic bunching")? HA: Failover is modest Failover times measured in seconds Reliable between hub correspondence DR: Failover is costly Failover times regularly measured in hours Unreliable between hub correspondence accepted Linux-HA gives extraordinary elements to manage "geographic bunching" (otherwise known as catastrophe recuperation)

Slide 14

Single Points of Failure (SPOFs) A solitary purpose of disappointment is a segment whose disappointment will bring about close quick disappointment of a whole framework or administration Good HA outline disposes of single purposes of disappointment

Slide 15

Non-Obvious SPOFs Replication connections are once in a while single purposes of disappointment The framework may fizzle when another disappointment happens Some plate controllers have SPOFs inside them which aren\'t evident without schematics Redundant connections covered in the same wire run have a typical SPOF Non-Obvious SPOFs can require profound ability to spot

Slide 16

The " Three R\'s " of High-Availability R edundancy R edundancy R edundancy If this sounds excess, that is most likely fitting... ;- ) Most SPOFs are dispensed with by excess HA Clustering is a decent method for giving and overseeing repetition

Slide 17

Redundant Communications Intra-group correspondence is basic to HA framework operation Most HA bunching frameworks give components to excess interior correspondence for heartbeats, and so on. Outside correspondences is normally crucial to arrangement of administration Exernal correspondence repetition is typically refined through directing traps Having a specialist in BGP or OSPF is an assistance

Slide 18

Fencing Guarantees asset honesty on account of certain troublesome cases Three Common Methods: FiberChannel Switch lockouts SCSI Reserve/Release (hard to make dependable) Self-Fencing (like IBM ServeRAID) STONITH – Shoot The Other Node In The Head Linux-HA bolsters the last two models

Slide 19

Data Sharing - None Strangely enough, some HA setups needn\'t bother with any formal plate information sharing Firewalls Load Balancers (Caching) Proxy Servers Static web servers whose substance is duplicated from a solitary source

Slide 20

Data Sharing – Replication Some applications give their own replication DNS, DHCP, LDAP, DB2, and so on. Linux has magnificent circle replication strategies accessible DRBD is my most loved DRBD-based HA bunches are greatly reasonable Some situations can live with less "exact" replication techniques – rsync, and so on. Frequently does not bolster parallel access Fencing profoundly alluring, but rather not generally vital EXTREMELY practical

Slide 22

Data Sharing – ServeRAID IBM ServeRAID circle is self-fencing This helps trustworthiness in failover situations This makes bunch filesystems, and so on incomprehensible No Oracle RAC, no GPFS, and so on. ServeRAID failover requires a script to perform volume handover Linux-HA gives such a script in open source Linux-HA is ServerProven with IBM ServeRAID

Slide 23

Data Sharing – FiberChannel The most great information sharing instrument Allows for failover mode Allows for genuine parallel access Oracle RAC, Cluster filesystems, and so forth. Fencing constantly required with FiberChannel Linux-HA is ensured ServerProven with IBM stockpiling

Slide 25

Data Sharing – Back-End Network Attached Storage can go about as an information sharing technique Existing Back End databases can likewise go about as an information sharing system Both make solid and repetitive information sharing Somebody Else\'s Problem (SEP). In the event that they benefited a vocation, you can profit by them. Be careful SPOFs in your neighborhood system

Slide 26

Linux-HA Background The most seasoned and most surely understood open-group HA venture - giving refined fall flat over and restart abilities for Linux (and different OSes) In presence since 1998; ~ 30k mission-basic bunches underway since 1999 Active, open advancement group drove by IBM and Novell Wide assortment of enterprises, applications upheld Shipped with most Linux circulations (everything except Red Hat) No extraordinary equipment necessities; no part conditions, all client space All discharges tried via robotized test suites

Slide 27

Linux-HA Capabilities Supports n-hub bunches – where "n" <= something like 16 Can utilize serial, UDP bcast, mcast, ucast comm. Fizzles over on hub disappointment, or on administration disappointment Fails over on loss of IP network, or self-assertive criteria Active/Passive or full Active/Active Built-in asset checking Support for the OCF asset standard Sophisticated reliance model with rich imperative bolster (assets, bunches, incarnations, expert/slave) (required for SAP) XML-based asset design Configuration and observing GUI Support for OCFS group filesystem Multi-state (expert/slave) asset bolster Split-site (stretch) bunch support with majority daemon

Slide 28

Some Linux-HA Terminology Node – a PC (genuine or virtual) which is a piece of the group and running our group programming stack Resource – something we oversee – an administration, or IP address, or circle drive, or whatever. On the off chance that we oversee it and it\'s not a hub, it\'s an asset Resource Agent – a script which goes about as an intermediary to control an asset. Most are firmly displayed after standard framework init scripts. DC – Designated Coordinator – the "expert hub" in the group STONITH – Acronym for Shoot The Other Node In The Head – a technique for fencing out hubs which are acting up by resetting them Partitioned bunch or Split-Brain – a condition where the group is part into two or more pieces which don\'t think about each other through equipment or programming disappointment. Kept from doing BadThings by STONITH Quorum – ordinarily allocated to at most one single parcel in a group to keep part cerebrum from bringing about harm. Commonly dictated by a voting convention

Slide 29

Key Linux-HA Processes CRM – Cluster Resource Manager – The principle administration substance in the group CIB – The bunch Information Base – attendant of data about assets, hubs. Additionally used to allude to the data oversaw by the CIB procedure. The CIB is XML-based. PE – Policy Engine – figures out what ought to be done given the present approach essentially – makes a diagram for the TE containing the things that should be done to align the bunch back with strategy (just keeps running on the DC) TE – Carries out the mandates made by the PE – through it\'s chart (just keeps running on the DC) CCM – Consensus Cluster Membership – figures out who is in the group, and who is most certainly not. A kind of watchman for group hubs. LRM – Local Resource Manager – low level process that does everything that requirements doing – not group mindful – no information of approach – eventually determined by the TE (through the different CRM forms) stonithd – daemon completing STONITH orders pulse – low level introduction and commun

View more...