36 lines
1.5 KiB
Plaintext
36 lines
1.5 KiB
Plaintext
---
|
|
summary: How should a fault-tolerant application service connect to Matrix?
|
|
---
|
|
created: 2015-04-12 13:56:23.0
|
|
creator: matt-williams
|
|
description: |-
|
|
I built a SIP/Matrix gateway. For carrier-grade SIP, we want to avoid single points of failure. How should a fault-tolerant SIP/Matrix gateway connect to Matrix?
|
|
|
|
It's possible to build an application service using an active/backup fault-tolerance architecture, where only one of the pair is connected into Matrix at a time (or both are connected but only one processes events).
|
|
|
|
I can't currently see a way of building an application service using a clustered architecture where many servers are all active but only a subset of these are required for capacity. Each instance in the cluster could connect to Matrix, but then events would be duplicated - we only need each event to be delivered to the cluster once.
|
|
|
|
The SIP approach (using multiple SRV records, one pointing to each host in the cluster) is outlined in RFC 3263. An alternative might be to allow application services to specify a "group" they're joining when they register.
|
|
id: '11330'
|
|
key: SYN-342
|
|
number: '342'
|
|
priority: '3'
|
|
project: '10000'
|
|
reporter: matt-williams
|
|
status: '10100'
|
|
type: '4'
|
|
updated: 2016-11-07 18:27:55.0
|
|
votes: '0'
|
|
watches: '2'
|
|
workflowId: '11430'
|
|
---
|
|
actions:
|
|
- author: richvdh
|
|
body: 'Migrated to github: https://github.com/matrix-org/synapse/issues/1308'
|
|
created: 2016-11-07 18:27:55.0
|
|
id: '13629'
|
|
issue: '11330'
|
|
type: comment
|
|
updateauthor: richvdh
|
|
updated: 2016-11-07 18:27:55.0
|