Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!


How to build an HA app/DB setup? - Page 2
New on LowEndTalk? Please Register and read our Community Rules.

All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.

How to build an HA app/DB setup?

2»

Comments

  • SplitIceSplitIce Member, Host Rep

    @deadbeef said:

    Latency will kill your writes.

    Gallera is designed for LAN not WAN. Replication is normally used for WAN. If you are going to use Galera over a WAN you are going to want to increase the lag time (measured in transactions), and limit your number of transactions.

    Thanked by 1deadbeef
  • raindog308raindog308 Administrator, Veteran

    vimalware said: You'll end up re-implementing Google Borg - > kubernetes :)

    Google Borg is not public :-) Kubernetes is container-oriented...most of my stuff wouldn't really be container-oriented (e.g., backups, seedboxes, image processing, etc.)

    But yes, in the future, I would like to think about these sorts of things...have a mesh of systems on top of which everything is containerized, can move wherever, scale up/down, etc. On the other hand, who am I kidding, I have LEBs.

    vimalware said: Next up: Mesos+/Marathon for job scheduling?

    The problem with these systems is they're big...Mesos is not going to be happy sharing MySQL on a 1GB box.

    BTW, Amazon Lambda now has scheduled events...

    Forgot to mention that I'm talking to the nodes over Amazon SQS because...I felt like it. Unless the queue polling chews bandwidth in which case I'll have the brains talk directly to a daemon on the node.

    deadbeef said: Latency will kill your writes.

    So another option is to put two in the same DC and then replicate on somewhere else - e.g., two in NY in sync, one somewhere else async. This gets into the greater messiness of active/passive failover. Percona requires 3 so the two peers would need to be something else.

    But @yomero is right - it's a job scheduling system and an inventory system. I definitely prefer availability over speed. The inventory side of it (CMDB) is read a ton but updated infrequently...the job scheduling is not super-write-heavy and the writes are not time-sensitive. Since I have etcd running, anything that might be latency-sensitive about the clustering there instead of the DB.

    Of course I'm just talking out my ass at the moment...I'll come back after a while and tell you all how I ignored your advice and what a dumbass I am.

    Thanked by 2rincewind deadbeef
  • raindog308raindog308 Administrator, Veteran

    SplitIce said: Gallera is designed for LAN not WAN. Replication is normally used for WAN. If you are going to use Galera over a WAN you are going to want to increase the lag time (measured in transactions), and limit your number of transactions.

    Yep:

    https://www.percona.com/blog/2016/03/14/percona-xtradb-cluster-in-a-high-latency-network-environment/

    I think I'll be somewhat higher...:-)

  • What's the point of windmills if you don't charge at them.

    Go full tilt, and enjoy the ride.

Sign In or Register to comment.