Add reader_vip_cidr

[djuntgen]

Add functionality to MHA-Helper to allow for reader_vip_cidr management.

• reader_vip_cidr= VIP for read only
• candidate_reader = boolean, if true then slave is eligible to take on reader vip.

MHA-Helper configuration

[default]
master_vip_cidr=192.168.4.10/24
reader_vip_cidr=192.168.4.11/24

[db1]
candidate_reader=1

[db2]
candidate_reader=1

# db3
[db3] 
candidate_reader=1

How it would work

Take db topology:

db1 (master holds master VIP)
  \ -> db2 (replica, holds reader VIP, candidate_master = 1)
  \ -> db3 (replica, none, candidate_reader = 1)

Case:

1. Make db2 master
2. Make db1 replica of db2
3. Move master_vip_cidr from db1 -> db2
4. Move reader_vip_cidr from db2 -> db3

There are some cases that need to be worked out like:

• What happens if db3 is down, or far behind in replication?
• What if db1 and db2 are the only master, replica pair, where does the reader vip go?
  • I think this is easy, just put the reader_vip on the master.

etc..

thoughts?

[ovaistariq]

Hi @djuntgen,

I have thought about this in past. I think it would be a good feature to have.
But for it to work, I think mha-helper would have to work as a daemon so that it is constantly monitoring the slave so that it can move the reader VIP when:

• the slave replication gets broken
• the slave is lagging more than a specific threshold
• the slave goes down

What do you think?

[djuntgen]

Thanks for review @ovaistariq,

I Agree with you, but I propose a phase approach to get basic functionality with current code base.

Requirements

• Must support more than one reader vip (maybe accomplished by executing two daemons)
• Must allow for monitoring and alert when reader vips move
• Configuration for when to move vips from one server to another
• Checks for split-brain and corrects

Phase 1

(dumb mode) - can be accomplished by modifying MHA-Helper's to execute move during master alive/failover execution.

• Allow for simply reader vip assignment to specified replica
• Looks for the best reader candidate from configuration
• Checks that replica isn't the master
• Checks that replication is running only

Phase 2

(smart mode) MHA-Helper has a secondary app

• runs as a daemon
• Looks for the best reader candidate from configuration
• Allows master to hold reader vips if no eligible replicas
• Monitors slave status
  • Checks that replica isn't the master
  • Checks that replication is running
  • Checks slave lag
• Allows for soft thresholds:
  • Replication running
  • Moves reader vips based on if replication hasn't ran for x seconds
  • Moves reader vips base on seconds behind master