Enhancement #625


mail: setup multi-master sync

Added by Marc Dequènes about 6 years ago. Updated almost 6 years ago.

Service :: Mail
Start date:
Due date:
% Done:


Estimated time:
Patch Available:
Help Needed:


We need to decide which host would be the second node. Maybe Toushirou-NG?

Related issues 1 (0 open1 closed)

Blocked by DuckCorp Infrastructure - Bug #435: DSPAM has no futureResolvedMarc Dequènes2015-02-04

Actions #1

Updated by Marc Dequènes about 6 years ago

  • Category set to Service :: Mail
Actions #2

Updated by Marc Dequènes almost 6 years ago

  • Blocked by Bug #435: DSPAM has no future added
Actions #3

Updated by Marc Dequènes almost 6 years ago

Because this would impact retraining using the recommended method by moving mails in/out of the Junk folder, we would first need to prepare the full anti* stack on the other node. I would rather replace DSPAM first (#435) than deploying it again. It is quite resource intensive for Toushirou to withhold it and I'm not sure two instance would not tramp on each-other while using the database.

Actions #4

Updated by Marc Dequènes almost 6 years ago

  • Status changed from New to In Progress
  • % Done changed from 0 to 40

~3 days ago I finished building the stack on Toushirou and worked on this replication setup. After some initial difficulties it seem to sync well. I know need to test using IMAP on Toushirou.

I deactivated #Virtual, only tested on my account and not really useful; it could not be replicated.

I added a filter to avoid #MilkyPond and #Shared replication which is not suppose to work either. No one seems to be using the features, so no real loss. I'll have a look if #MilkyPond could be restored just for fun.

Before I can bring this to production I need to Ansibilize Spoolinger. I did some code improvement and added a service file and plan to package it. Then retraining can be activated too on Toushirou.

Actions #5

Updated by Marc Dequènes almost 6 years ago

  • % Done changed from 40 to 50

Spoolinger is packaged and Ansibilized now. It was deployed on Toushirou.

Now I need to test interacting with both servers at the same time and see how synchronization and retraining behaves.

If the use triggers a retraining on one server and cancel it on the other, this would not work well. There's no easy way to fix this but I think cancellation happens really rarely and on the same session so we can forget about it.

Actions #6

Updated by Marc Dequènes almost 6 years ago

  • Status changed from In Progress to Resolved
  • % Done changed from 50 to 100

IMAP and SIEVE are working fine and replication is very fast.
I setup the delivery with mailman routing to Orfeo.
Toushirou is a full MX1 now.


Also available in: Atom PDF