-
Notifications
You must be signed in to change notification settings - Fork 2
- The failure of an instance in the cloud is fine and the services take over
- Command like starting, creating instances, auto provisionning will be disable
This is set like this for security reasons, the running cloud instances does not have a clue about the your cloud password thos information stay in your office or at home. Still it is possible to run multiple private instances for no single point of failure in your office.
Warning : the Public Keys need to be propagated
ScrambleDB use 4 of them in the framework but a plan to have one more level could happen in futur realeases, each proxy have a specific role:
Keepalived maintain the VIP and report to the cluster who is the active director. In the cloud this looks unecesary but for internal network some decisions like promoting a new memcache or persistente store must be faster compare to the speed of any cloud API.
Ha-Proxy is used for load balancing and can auto remove dead services as well holding multiple traffic load balancing.
MySQL-proxy is the the SQL routing manager it is reliable and fast enouth for managing the MySQL protocol.
Spider is the proxy in charge of sharding it use 2 phase commit to offer a centralize master view of the data.
We will expose that parameter for every services in the configuration file but for now: In lib/Scramble/Common/ClusterUtils.pm
Change the verbosity our $g_log_level=2;
For web servers, proxying and Key value store, it should be trivial to make good use of it. For Storage node you need to make sure that replication can be made from a CDC service a good way to achieve this is using drizzle as a CDC slave + replication plugins to update memcache GTTID and apply to any non MySQL slave.