Open shawninder opened 8 years ago
Thank you for opening this, i have been meaning to do it in a while, just haven't had time.
On 10 February 2016 at 15:13, Shawn Inder notifications@github.com wrote:
see vigour-io/packer-server#31 https://github.com/vigour-io/packer-server/issues/31
— Reply to this email directly or view it on GitHub https://github.com/vigour-io/git-spy/issues/6#issuecomment-182391875.
@andrepadez Have you thought of the architecture for this?
I think a hub would be a perfect space in which git-spy and packer can communicate.
It's fast, works no matter where each process is actually located (different machines or not), and message passing works even if one of the services is down momentarily.
yes, i thought about something like that. But if we want this fast, we will have to implement a pure websocket solution
On 10 February 2016 at 17:12, Shawn Inder notifications@github.com wrote:
@andrepadez https://github.com/andrepadez Have you thought of the architecture for this?
I think a hub would be a perfect space in which git-spy and packer can communicate.
- git-spy makes sure to put whatever data the packers need in the hub (commit hashes, diffs, whatever) by poking the GitHub API at startup and keeps it up to date by listening for webhooks from GitHub
- packer-server knows from the hub what to do at startup and listens for changes on that data
It's fast, works no matter where each process is actually located (different machines or not), and message passing works even if one of the services is down momentarily.
— Reply to this email directly or view it on GitHub https://github.com/vigour-io/git-spy/issues/6#issuecomment-182451618.
Is the hub not stable enough?
I don't think so
On 10 February 2016 at 18:17, Shawn Inder notifications@github.com wrote:
Is the hub not stable enough?
— Reply to this email directly or view it on GitHub https://github.com/vigour-io/git-spy/issues/6#issuecomment-182488694.
There is a maximum number of webhooks (20) we can wet on an organization. We will bust that limit pretty soon at this rate. I think the solution is to stop having one git-spy per packer, but rather having a single git-spy to which many packers can subscribe.
This means adding a way to subscribe to git-spy without having to require it (web sockets? inter-process messaging? https?)