| Internet-Draft | IGP for Network HA | April 2023 |
| Chen, et al. | Expires 8 October 2023 | [Page] |
This document describes protocol extensions to OSPF and IS-IS for improving the reliability or availability of a network controlled by a controller cluster.¶
The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in RFC 2119 [RFC2119].¶
This Internet-Draft is submitted in full conformance with the provisions of BCP 78 and BCP 79.¶
Internet-Drafts are working documents of the Internet Engineering Task Force (IETF). Note that other groups may also distribute working documents as Internet-Drafts. The list of current Internet-Drafts is at https://datatracker.ietf.org/drafts/current/.¶
Internet-Drafts are draft documents valid for a maximum of six months and may be updated, replaced, or obsoleted by other documents at any time. It is inappropriate to use Internet-Drafts as reference material or to cite them other than as "work in progress."¶
This Internet-Draft will expire on 8 October 2023.¶
Copyright (c) 2023 IETF Trust and the persons identified as the document authors. All rights reserved.¶
This document is subject to BCP 78 and the IETF Trust's Legal Provisions Relating to IETF Documents (https://trustee.ietf.org/license-info) in effect on the date of publication of this document. Please review these documents carefully, as they describe your rights and restrictions with respect to this document. Code Components extracted from this document must include Revised BSD License text as described in Section 4.e of the Trust Legal Provisions and are provided without warranty as described in the Revised BSD License.¶
More and more networks are controlled by central controllers or controller clusters. A controller cluster is a single controller externally. It normally consists of two or more controllers internally working together to control a network, i.e., every network element (NE) in the network. The reliability or availability of a network is heavily dependent on its controller cluster. The issues or failures in the controller cluster may impact the reliability or availability of the network greatly.¶
For a controller cluster comprising two or more controllers (i.e., primary controller, secondary controller, and so on), the failures in the cluster may split the cluster into a few of separated controller groups. These groups do not know each other and may be out of synchronization. Two or more groups may be elected to control the network at the same time, which may cause some issues.¶
This document proposes some procedures and extensions to OSPF and IS-IS for the separated controllers or controller groups to know each other thus elect one new primary controller or controller group correctly when the cluster is split because of failures in the cluster.¶
The following terminologies are used in this document.¶
This section briefs the mechanism of controller cluster reliability or availability using IGP, and illustrates some details through a simple example.¶
When a cluster of controllers is split into a few of separated groups because of failures in the cluster, the live controllers are still actually connected to the network (i.e., network elements). Through some of these connections, each group can get the information about the other groups. A new primary controller or controller group is correctly elected to control the network based on the information.¶
Each controller may comprise an IGP as an information proxy, called IGP information proxy or IGP for short. The IGP has an IGP adjacency relation with each of a given number of NEs (such as one NE) in the network. When one adjacency is broken, a new adjacency is created and maintained if possible. The given number of adjacency relations is retained.¶
In normal operations, the cluster has all its controllers connected. They are the primary controller controlling the network, the secondary controller, and so on. They have current position 1, 2, and so on respectively. The primary controller advertises the information about the controllers via its IGP adjacencies. The extensions to IGP below is used.¶
When the cluster is split into a few separated groups, each group elects an intent primary controller, secondary controller and so on from the group, which have intent position 1, 2, and so on respectively. The intent primary controller advertises the information about the controllers in the group.¶
The information advertised by the (intent) primary controller includes its current (intent) position, its old position, its priority to become a primary controller, the number of controllers, and the IDs of the controllers which are ordered according to their (intent) positions. In addition, a flag C indicating that whether it is Controlling the network (i.e., it is the primary controller or intent primary controller) is included.¶
Figure 1 shows a controller cluster comprising two controllers: the primary controller and the secondary controller. Each controller includes an IGP as an information proxy.¶
+---------------------------------------------------+
| Controller Cluster |
| |
| +------------+ +------------+ |
| |Controller A| Synchronize |Controller B| |
| |(Primary) +---------------+(Secondary) | |
| | [IGP]| | [IGP]| |
| ++-----------+ +-----------++ |
| | ^ | |
| | |_______________ | |
| | | | |
| | v | |
+-----|------------Control Channels-----------|-----+
| / \ |
|IGP Adj / \____ |
\ / \ \____ |IGP Adj
\____ /\ .---. .---+ \ |
\ | \( ' |'.---. | |
\ |---\ Network | '+. |
NE1 (o \ | | ) /
( | | o) NE4
( | | )
( o NE2 o NE3.-'
' )
'---._.-. )
'---'
The IGP in a controller has one IGP adjacency relation with one NE in the network. In Figure 1, the IGP in controller A has IGP adjacency with NE1, the IGP in B has IGP adjacency with NE4.¶
In normal operations, the IGP of the primary controller originates link state (LS) containing the information about the controllers connected to it. The LS originated by Controller A (Primary) in Figure 1 having the following contents:¶
C = 1, A's current Position = 1, A's OldPosition = 1, A's Priority, NoControllers = 2, A's ID, B's ID¶
When failures happen in the cluster, the live controllers act as follows:¶
For the Secondary Controller (e.g., B) alive, if the primary controller is dead, it promotes itself as the new primary controller; if the primary controller is alive but separated from the secondary controller, the secondary controller will not promote itself to be a new primary controller.¶
For the Primary Controller (e.g., A), if it is alive, it continues to be the primary controller.¶
With the extensions to IGP, the secondary controller can determine the status of the primary controller through using IGP and obtaining the information about the primary controller. The conditions that the primary controller is alive but separated from the secondary controller (i.e., condition a: the connection between the primary controller and the secondary controller in the cluster failed, but condition b: the two controllers are alive) can be determined by the secondary controller as follows:¶
For condition a, when the heartbeat from the primary stops, the secondary knows that the connection between the primary and secondary controller failed.¶
For condition b, it checks its link state database (LSDB) in the IGP to see whether the IGP for the primary controller is connected to some network elements and advertises the LS. If so, the primary controller is alive; otherwise, it is dead.¶
This section describes extensions to OSPF and IS-IS.¶
A new TLV, called OSPF Controllers TLV, is defined. When OSPF acts as a proxy of a controller in a cluster, it may advertise the information about the controllers such as the number of controllers connected to it (including itself) in its router information LSA, which contains a Controllers TLV of the following format.¶
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type (TBD1) | Length | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Flags |C| Position | OldPosition | Priority | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Reserved | NoControllers | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Controller 1 ID | : : | | Controller n ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
When the information about the controllers is changed, OSPF of a primary controller originates an OSPF Router Information Opaque LSA, which includes a OSPF Controllers TLV.¶
Similar to OSPF, a new TLV, called IS-IS Controllers TLV, is defined. When IS-IS acts as a proxy of a controller in a cluster, it may advertise the information about the cluster such as the number of controllers connected to it (including itself) in its LSP, which contains an IS-IS Controllers TLV of the following format.¶
0 1 2 3 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 2 3 4 5 6 7 8 9 0 1 +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Type (TBD2) | Length | Flags |C| Position | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | OldPosition | Priority | NoControllers | Reserved | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+ | Controller 1 ID | : : | | Controller n ID | +-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+-+
When the information about the controllers is changed, the IS-IS of a primary controller originates an LSP, which includes an IS-IS Controllers TLV.¶
This section describes the recovery procedure for a controller cluster of n (n > 2) controllers, which are the primary controller A, the secondary controller B, ..., the n-th controller N.¶
When failures happen in the cluster, it may be split into a few separated groups of controllers. In one policy, the group with the maximum number of controllers is responsible for controlling the network as the primary group of the cluster, in which the new primary controller, secondary controller, and so on are elected.¶
For each separated group of controllers, the intent primary controller, secondary controller, and so on are elected. The intent primary controller of the group advertises the information about the group through its IGP. The information includes its intent position, its old position, its priority to become a primary controller, the number of controllers in the group, and identifiers of the controllers in the group. The identifiers of the controllers are ordered according to their positions. The identifier of the intent primary controller, which has position 1, is the first one; The identifier of the intent secondary controller, which has position 2, is the second one; and so on. Thus every separated group has the information about the other groups and can determine which group has the maximum number of controllers.¶
In the case of tie (i.e., two or more groups have the same maximum number of controllers), the group with the highest priority controller wins in one policy. In another policy, the group with the highest old position controller (e.g., the old primary controller) wins.¶
Some details of the recovery procedures in the current and intent primary controller in a controller cluster or group are as follows.¶
In normal operations, it advertises Controllers TLV containing:¶
C = 1, Position = 1, Old Position = 1, Primary Controller's priority, NoControllers = n, Primary Controller's ID, secondary controller's ID, ..., and n-th Controller's ID.¶
When failures cause the cluster split, it advertises Controllers TLV containing:¶
C = 0, Position = 1, Old Position = 1, Intent Primary Controller's priority, NoControllers = m (m is the number of controllers in the group that the primary controller is connected after the failures), Intent Primary Controller's ID, IDs of the other controllers connected.¶
Then after a given time, it checks if the group is elected as the primary group. If so, it advertises Controllers TLV containing:¶
C = 1, Position = 1, Old Position = 1, its Priority, NoControllers = m, the IDs of the controllers in the group.¶
One example is that failures split the cluster into two separated groups: group 1 comprising A and C, group 2 consisting of B and N. Each group elects its intent primary controller, secondary controller, and so on. Suppose that controller A and C are elected as the intent primary and secondary controller respectively in group 1; controller B and N are elected as the intent primary and secondary controller respectively in group 2.¶
Each of the intent primary controllers A and B advertises the information about the controllers in its group. The information advertised by A includes:¶
C = 0, Position = 1, OldPosition = 1, A's Priority, NoControllers = 2, A's ID, C's ID.¶
The information advertised by B includes:¶
C = 0, Position = 1, OldPosition = 2, B's Priority, NoControllers = 2, B's ID, N's ID.¶
Group 1 and 2 have the same number of controllers, which is 2. But OldPosition in group 1 is higher than that in group 2. Group 1 is elected as the primary group, and the intent primary controller A in the primary group is determined as the current primary controller. After the determination, the information about the controllers in group 1 (i.e., the primary group) is changed. The updated information advertised by A includes:¶
C = 1, Position = 1, OldPosition = 1, A's Priority, NoControllers = 2, A's ID, C's ID.¶
TBD¶