From 15ec6030ffccdbddc06551039d9ebf6b86a5cfd1 Mon Sep 17 00:00:00 2001
From: Spike Curtis <spike@coder.com>
Date: Tue, 15 Oct 2024 14:05:27 +0400
Subject: [PATCH] chore: add http/pprof server over unix socket for debug
 (#295) (#296)

* chore: add http/pprof server over unix socket for debug



* remove old pprof file without checking if it exists



---------

Signed-off-by: Spike Curtis <spike@coder.com>
---
 main.go          |  1 +
 pprof_unix.go    | 42 ++++++++++++++++++++++++++++++++++++++++++
 pprof_windows.go |  6 ++++++
 3 files changed, 49 insertions(+)
 create mode 100644 pprof_unix.go
 create mode 100644 pprof_windows.go

diff --git a/main.go b/main.go
index 000436a3..a16c9951 100644
--- a/main.go
+++ b/main.go
@@ -11,6 +11,7 @@ import (
 //go:generate go run github.com/hashicorp/terraform-plugin-docs/cmd/tfplugindocs
 
 func main() {
+	servePprof()
 	plugin.Serve(&plugin.ServeOpts{
 		ProviderFunc: provider.New,
 	})
diff --git a/pprof_unix.go b/pprof_unix.go
new file mode 100644
index 00000000..717bc01b
--- /dev/null
+++ b/pprof_unix.go
@@ -0,0 +1,42 @@
+//go:build !windows
+
+package main
+
+import (
+	"net"
+	"net/http"
+	"net/http/pprof"
+	"os"
+)
+
+// servePprof starts an HTTP server running the pprof goroutine handler on a local unix domain socket. As described in
+// https://github.com/coder/coder/issues/14726 it appears this process is sometimes hanging, unable to exit cleanly,
+// and this prevents additional Coder builds that try to reinstall this provider.  A goroutine dump should allow us to
+// determine what is hanging.
+//
+// This function is best-effort, and just returns early if we fail to set up the directory/listener. We don't want to
+// block the normal functioning of the provider.
+func servePprof() {
+	// Coder runs terraform in a per-build subdirectory of the work directory.  The per-build subdirectory uses a
+	// generated name and is deleted at the end of a build, so we want to place our unix socket up one directory level
+	// in the provisionerd work directory, so we can connect to it from provisionerd.
+	err := os.Mkdir("../.coder", 0o700)
+	if err != nil && !os.IsExist(err) {
+		return
+	}
+
+	// remove the old file, if it exists. It's probably from the last run of the provider
+	if err = os.Remove("../.coder/pprof"); err != nil && !os.IsNotExist(err) {
+		return
+	}
+	l, err := net.Listen("unix", "../.coder/pprof")
+	if err != nil {
+		return
+	}
+	mux := http.NewServeMux()
+	mux.Handle("/debug/pprof/goroutine", pprof.Handler("goroutine"))
+	srv := http.Server{Handler: mux}
+	go srv.Serve(l)
+	// We just leave the server and domain socket up forever. Go programs exit when the `main()` function returns, so
+	// this won't block exiting, and it ensures the pprof server stays up for the entire lifetime of the provider.
+}
diff --git a/pprof_windows.go b/pprof_windows.go
new file mode 100644
index 00000000..05cd3143
--- /dev/null
+++ b/pprof_windows.go
@@ -0,0 +1,6 @@
+//go:build windows
+
+package main
+
+// servePprof is not supported on Windows
+func servePprof() {}