mirror of
https://github.com/zap-proto/zip.git
synced 2026-10-19 15:38:57 +00:00
zeekayandHanzo Dev
9118f9c2e4
The verb deletion is done and the library builds: Graft, Add, Service, the Mount VERB and the claims ledger are gone; Load and Mount return definitions; Plugins/Reload/Unload now find their plugin by walking. NOT DONE — the same entanglement, one layer over: a plugin's MCP catalogue (pluginTools/toolOwners/open) is still read off the host that loaded it, but a plugin is now a definition, so the host reads an empty set and serves no door. Four mcp_compose/mcp_open tests fail on exactly that. The fix is the same one that worked for Plugins(): aggregate across App.hosts() instead of reading a private map, and move the duplicate-tool-name refusal to the walk so two plugins can actually see each other. Co-authored-by: Hanzo Dev <[email protected]>
64 lines
2.1 KiB
Go
64 lines
2.1 KiB
Go
package zip_test
|
|
|
|
// What a plugin costs per request.
|
|
//
|
|
// Running a service as its own binary buys independent build, release, restart
|
|
// and measurement — and charges a hop for it. This benchmark prices that hop by
|
|
// serving the SAME route two ways through the same host: once by a handler
|
|
// linked into the host, once by a child process the host reaches over its unix
|
|
// socket. Everything else is held equal, so the delta IS the hop.
|
|
//
|
|
// It exists because the trade is only decidable with the number in hand. A hop
|
|
// that is small against a real handler's work is a deployment decision; one that
|
|
// is large is an architecture constraint, and the two look identical in prose.
|
|
//
|
|
// Run one round per process and take the minimum across rounds, so A and B
|
|
// interleave under drifting load instead of A running entirely before B:
|
|
//
|
|
// for i in $(seq 8); do
|
|
// go test -run='^$' -bench=Benchmark_PluginHop -benchmem -count=1 -benchtime=3000x .
|
|
// done
|
|
//
|
|
// Numbers and methodology: BENCHMARKS.md §3.
|
|
|
|
import (
|
|
"testing"
|
|
|
|
"github.com/zap-proto/zip"
|
|
)
|
|
|
|
// Benchmark_PluginHop compares a linked-in route against the same route served
|
|
// by a loaded plugin. Both dispatch through app.Fiber().Handler() on a reused
|
|
// *fasthttp.RequestCtx, the same idiom as Benchmark_ZipTax, so neither side pays
|
|
// net/http↔fasthttp conversion.
|
|
func Benchmark_PluginHop(b *testing.B) {
|
|
bin := buildPlugin(b, "v1")
|
|
|
|
// The linked-in baseline answers byte-for-byte what the plugin answers, so
|
|
// the only difference under test is where the handler runs.
|
|
linked := zip.New(benchConfig())
|
|
linked.Get("/v1/demo/version", func(c *zip.Ctx) error {
|
|
return c.JSON(200, map[string]string{"version": "v1"})
|
|
})
|
|
|
|
host := zip.New(benchConfig())
|
|
host.Use(must(zip.Load(zip.Plugin{Name: "demo", Bin: bin}, "/v1/demo")))
|
|
defer func() { _ = host.Shutdown() }()
|
|
|
|
for _, tc := range []struct {
|
|
name string
|
|
app *zip.App
|
|
}{{"linked", linked}, {"plugin", host}} {
|
|
h := tc.app.Fiber().Handler()
|
|
b.Run(tc.name, func(b *testing.B) {
|
|
fctx := getFctx("/v1/demo/version")
|
|
b.ReportAllocs()
|
|
b.ResetTimer()
|
|
for i := 0; i < b.N; i++ {
|
|
fctx.ResetUserValues()
|
|
h(fctx)
|
|
}
|
|
})
|
|
}
|
|
}
|