replication #30
replication #30
Conversation
We want to put all the replica files in the same directory so it makes sense just to specify a directory for storage and let the program figure out what to do from there.
Want to adjust how this is orchestrated but right now I see data going into both files
| @@ -1,6 +1,8 @@ | |||
| package client | |||
|
|
|||
| import "distributed/pkg/storage" | |||
| import ( | |||
| "distributed/pkg/server/storage" | |||
There was a problem hiding this comment.
might want to make client key/values.
| orchestrationLogger.Printf("error closing socket %s: %v", socket, err) | ||
| } | ||
| } | ||
| wg.Done() |
There was a problem hiding this comment.
Don't think I need this anymore now that I have Router.Run(). had it installed when the orchestrator wasn't waiting for client connections.
gmkohler
force-pushed
the
gmkohler/replication
branch
from
215837b
to
2a23076
Compare
|
Should I be able to just run cmd/orchestrator/main.go with nothing else running first? Looks like I get an error trying to dial /tmp/distkv/orchestration.sock and I don't see a prior Listen |
|
@ozan Seeing the same issue when running this from the root. In particular spawning the servers is failing: I had been relying on being in orchestration's directory. I'll make this easier to run and add a bit of docs about how to run and how the coordination works before requesting another view. regarding:
I had a server process open the socket and the orchestrator connect to it, because the orchestrator was meant to send messages to the server to accept primaries, but perhaps I'm too tied to the web request/response pattern here and can have the orchestrator wait for the (primary) server to connect to a socket the orchestrator opened. |
Idea is to launch the
cmd/orchestrator, which will spawn a primary and a replica and run the routing layer. Then you can connect to the orchestrator withcmd/clientSeeking feedback on general design and whether I'm being naïve about interfacing with the sockets.