Bug: Multiple Processes Per Plugin

[tooolbox]

I wrote a Go plugin modeled precisely on your example.

Based on my testing, plugins are launched as processes and are never shut down, meaning a new instance is spawned every time a plugin function is called, and the old instances don't shut down. I noticed you have a concern about data races surrounding shutdown?

Spawning infinite plugin instances is not an option for me. Can this be fixed?

[xyproto]

Thanks for reporting! I will look into this.

[tooolbox]

Great, no problem. If it makes a difference, I'm running on Windows.

In the meantime I have hacked it by making the plugin process kill itself after a few seconds.

[xyproto]

Part of the idea of the plugin system, as opposed to just running an executable, is that the plugin can keep running and continue to respond as calls to the plugin is made.

See also: https://github.com/natefinch/pie#about-pie

So, the plugin executables should ideally not be closed per call, but there shouldn't be a new one started per call either (so that's a bug).

[tooolbox]

My one question (docs point) would be what the expected actions would be when updating or replacing plugins, in the case that they're long-running. I guess you restart Algernon in that case?

[xyproto]

Yes, just like with FastCGI, where executables keep running, I would think that restarting the server is what's required.

[tooolbox]

I never actually got this fixed. The plugin process keeps running in Windows even when PluginCode is called without a second argument.

Seems like this is actually the problem: natefinch/pie#14