Ceph

h1. RGW - ActiveActive Arch

h3. Summary

Current multizone RGW setup is only master-slave. In a single region we can only have a single master zone, and all data will be replicated to other slave zones. We plan to make it possible to change that behavior and have all zones in a single region writable. Also, we'll probably rename Region to Zone-Group due to some confusion.

h3. Owners

* Yehuda (Red Hat)

h3. Interested Parties

* Name (affiliation)

h3. Current Status

One master region, zero or more slave regions.

For each region, one master zone, zero or more secondary zones.

Data sync only happen from master zone to secondary zones

Metadata updates happen on master region

h3. Detailed Description

Requirements:

The ability to write data on any zone within a single region, have the data sync happen between all zones (in that region)

No change in metadata propagation

Proposal:

still designate a master zone for metadata changes

all zones within a single region follow all other zones for data sync

all zones within a single region will allow write (unless configured otherwise)

sync agent per zone will follow all relevant zones

bucket index log will contain source zone, and source bucket index log id, for each operation

replica log will hold a list of buck index log states for each bucket instance, one per zone (instead of a single state for the master zone)

extend system user copy operation that is used by the sync agent to be able to set source bucket index log op id for each op

sync agent will set bucket index op id on each object sync operation

extend rgw copy operation to be able to do conditional copy that will check target object modification time (copy if older)

sync agent will use the new conditional copy option

h3. Work items

h4. Coding tasks

# Task 1

# Task 2

# Task 3

h4. Build / release tasks

# Task 1

# Task 2

# Task 3

h4. Documentation tasks

# Task 1

# Task 2

# Task 3

h4. Deprecation tasks

# Task 1

# Task 2

# Task 3