Revision 60c6bb5e docs/admin-guide.rst
b/docs/admin-guide.rst | ||
---|---|---|
2536 | 2536 |
All sections: :ref:`Scale out Guide <i-synnefo>` |
2537 | 2537 |
|
2538 | 2538 |
|
2539 |
Regions, Zones and Clusters |
|
2540 |
=========================== |
|
2541 |
|
|
2542 |
Region |
|
2543 |
------ |
|
2544 |
|
|
2545 |
A Region is a single Synnefo installation, with |
|
2546 |
Compute/Network/Image/Volume/Object Store services. A Region is associated with |
|
2547 |
one set of Synnefo DBs (Astakos DB, Pithos DB and Cyclades DB). Every Region has a |
|
2548 |
distinct set of API endpoints, e.g., |
|
2549 |
`https://cloud.example.com/cyclades/compute/v2.0`. Two Regions are most times |
|
2550 |
located geographically far from each other, e.g. "Europe", "US-East". A Region |
|
2551 |
comprises multiple Zones. |
|
2552 |
|
|
2553 |
Zone |
|
2554 |
---- |
|
2555 |
|
|
2556 |
A Zone is a set of Ganeti clusters, in a potentially geographically distinct |
|
2557 |
location, e.g. "Athens", "Rome". All clusters have access to the same physical |
|
2558 |
networks, and are considered a single failure domain, e.g., they access the |
|
2559 |
network over the same router. A Zone comprises muliple Ganeti clusters. |
|
2560 |
|
|
2561 |
Ganeti cluster |
|
2562 |
-------------- |
|
2563 |
|
|
2564 |
A Ganeti cluster is a set of Ganeti nodes (physical machines). One of the nodes |
|
2565 |
has the role of "Ganeti master". If this node goes down, another node may |
|
2566 |
undertake the master role. Ganeti nodes run Virtual Machines (VMs). VMs can live |
|
2567 |
migrate inside a Ganeti cluster. A Ganeti cluster comprises multiple physical |
|
2568 |
hardware nodes, most times geographically close to each other. |
|
2569 |
|
|
2570 |
VM mobility |
|
2571 |
----------- |
|
2572 |
|
|
2573 |
VMs may move across Regions, Zones, Ganeti clusters and physical nodes. Before we |
|
2574 |
describe how that's possible, we will describe the different kinds of moving, |
|
2575 |
providing the corresponding terminology: |
|
2576 |
|
|
2577 |
Live migration |
|
2578 |
~~~~~~~~~~~~~~ |
|
2579 |
|
|
2580 |
The act of moving a running VM from physical node to physical node without any |
|
2581 |
impact on its operation. The VM continues to run on its new physical location, |
|
2582 |
completely unaffected, and without any service downtime or dropped connections. |
|
2583 |
Live migration typically requires shared storage and networking between the source |
|
2584 |
and destination nodes. |
|
2585 |
|
|
2586 |
Live migration is issued by the administrator in the background and is transparent |
|
2587 |
to the VM user. |
|
2588 |
|
|
2589 |
Failover |
|
2590 |
~~~~~~~~ |
|
2591 |
|
|
2592 |
The act of moving a VM from physical node to physical node by stopping it first on |
|
2593 |
the source node, then re-starting it on the destination node. There is short |
|
2594 |
service downtime, during the time the VM boots up, and client connections are |
|
2595 |
dropped. |
|
2596 |
|
|
2597 |
Failover is issued by the administrator in the background and the VM user will |
|
2598 |
experience a reboot. |
|
2599 |
|
|
2600 |
Snapshot Failover |
|
2601 |
~~~~~~~~~~~~~~~~~ |
|
2602 |
|
|
2603 |
The act of moving a VM from physical node to physical node via a point-in-time |
|
2604 |
snapshot. That is, stopping a VM on the source node, taking a snapshot, then |
|
2605 |
creating a new VM from that snapshot. |
|
2606 |
|
|
2607 |
Snapshot failover is issued by the VM user and not the administrator. |
|
2608 |
|
|
2609 |
Disaster Recovery |
|
2610 |
----------------- |
|
2611 |
|
|
2612 |
In Synnefo terminology, Disaster Recovery is the process of sustaining a disaster |
|
2613 |
in one datacenter, and ensuring business continuity by performing live migration |
|
2614 |
or failover of running/existing VMs, or respawning VMs from previously made |
|
2615 |
snapshots. Based on the method used, this can work inside a single Ganeti cluster, |
|
2616 |
across Ganeti clusters in the same Zone, or across Zones. |
|
2617 |
|
|
2618 |
Specifically: |
|
2619 |
|
|
2620 |
Live migration is only supported inside a single Ganeti cluster. Ganeti supports |
|
2621 |
live migration between nodes in the same cluster with or without shared storage. |
|
2622 |
Live migration is done at the Ganeti level and is transparent to Synnefo. |
|
2623 |
|
|
2624 |
Failover is supported inside a Ganeti cluster, across Ganeti clusters and across |
|
2625 |
Zones. Ganeti supports failover inside a Ganeti cluster with or without shared |
|
2626 |
storage, which poses minimum downtime for the VM. Failover inside the same Ganeti |
|
2627 |
cluster is done at the Ganeti level and is transparent to Synnefo. |
|
2628 |
|
|
2629 |
Ganeti also provides tools for failing over VMs across different Ganeti clusters, |
|
2630 |
meaning that one can use them to failover VMs across Ganeti clusters of the same |
|
2631 |
Zone or across Ganeti clusters of different Zones, thus moving across Zones. |
|
2632 |
Failing over across different Ganeti clusters requires copying of data, resulting |
|
2633 |
in longer downtimes, depending on the geographical distance and network between |
|
2634 |
them. Failover across Ganeti clusters, either in the same or different Zones, is |
|
2635 |
not transparent to Synnefo and requires manual import of intances at Synnefo level |
|
2636 |
too, by the administrator. |
|
2637 |
|
|
2638 |
Snapshot failover supports moving VMs across all domains. It is issued by the VM |
|
2639 |
user and is done at the Synnefo level without the need of running anything at the |
|
2640 |
Ganeti level or by the administrator. |
|
2641 |
|
|
2642 |
In the future Synnefo will also support moving VMs across different Regions. |
|
2643 |
|
|
2644 |
|
|
2539 | 2645 |
Upgrade Notes |
2540 | 2646 |
============= |
2541 | 2647 |
|
Also available in: Unified diff