From 74679e9e2efeb8956132a79bea16728d76a7fe5e Mon Sep 17 00:00:00 2001
From: Giuliano Mega <giuliano.mega@gmail.com>
Date: Mon, 11 Mar 2024 09:31:17 -0300
Subject: [PATCH] abort Codex when startup sequence fails (#738)

---
 codex.nim       | 40 ++++++++++++++++++++++++----------------
 codex/codex.nim |  6 ------
 2 files changed, 24 insertions(+), 22 deletions(-)

diff --git a/codex.nim b/codex.nim
index f7215db6..a262bbc8 100644
--- a/codex.nim
+++ b/codex.nim
@@ -80,7 +80,7 @@ when isMainModule:
 
     var
       state: CodexStatus
-      pendingFuts: seq[Future[void]]
+      shutdown: Future[void]
 
     let
       keyPath =
@@ -93,6 +93,12 @@ when isMainModule:
       server = CodexServer.new(config, privateKey)
 
     ## Ctrl+C handling
+    proc doShutdown() =
+      shutdown = server.stop()
+      state = CodexStatus.Stopping
+
+      notice "Stopping Codex"
+
     proc controlCHandler() {.noconv.} =
       when defined(windows):
         # workaround for https://github.com/nim-lang/Nim/issues/4057
@@ -101,10 +107,7 @@ when isMainModule:
         except Exception as exc: raiseAssert exc.msg # shouldn't happen
       notice "Shutting down after having received SIGINT"
 
-      pendingFuts.add(server.stop())
-      state = CodexStatus.Stopping
-
-      notice "Stopping Codex"
+      doShutdown()
 
     try:
       setControlCHook(controlCHandler)
@@ -116,26 +119,31 @@ when isMainModule:
       proc SIGTERMHandler(signal: cint) {.noconv.} =
         notice "Shutting down after having received SIGTERM"
 
-        pendingFuts.add(server.stop())
-        state = CodexStatus.Stopping
-
-        notice "Stopping Codex"
+        doShutdown()
 
       c_signal(ansi_c.SIGTERM, SIGTERMHandler)
 
-    pendingFuts.add(server.start())
+    try:
+      waitFor server.start()
+    except CatchableError as error:
+      error "Codex failed to start", error = error.msg
+      # XXX ideally we'd like to issue a stop instead of quitting cold turkey,
+      #   but this would mean we'd have to fix the implementation of all
+      #   services so they won't crash if we attempt to stop them before they
+      #   had a chance to start (currently you'll get a SISGSEV if you try to).
+      quit QuitFailure
 
     state = CodexStatus.Running
     while state == CodexStatus.Running:
       # poll chronos
       chronos.poll()
 
-    # wait fot futures to finish
-    let res = waitFor allFinished(pendingFuts)
-    state = CodexStatus.Stopped
-
-    if res.anyIt( it.failed ):
-      error "Codex didn't shutdown correctly"
+    try:
+      # signal handlers guarantee that the shutdown Future will
+      # be assigned before state switches to Stopping
+      waitFor shutdown
+    except CatchableError as error:
+      error "Codex didn't shutdown correctly", error = error.msg
       quit QuitFailure
 
     notice "Exited codex"
diff --git a/codex/codex.nim b/codex/codex.nim
index c5bcb7b6..637fbca6 100644
--- a/codex/codex.nim
+++ b/codex/codex.nim
@@ -46,7 +46,6 @@ logScope:
 
 type
   CodexServer* = ref object
-    runHandle: Future[void]
     config: CodexConf
     restServer: RestServerRef
     codexNode: CodexNodeRef
@@ -183,9 +182,6 @@ proc start*(s: CodexServer) {.async.} =
   await s.codexNode.start()
   s.restServer.start()
 
-  s.runHandle = newFuture[void]("codex.runHandle")
-  await s.runHandle
-
 proc stop*(s: CodexServer) {.async.} =
   notice "Stopping codex node"
 
@@ -196,8 +192,6 @@ proc stop*(s: CodexServer) {.async.} =
     s.repoStore.stop(),
     s.maintenance.stop())
 
-  s.runHandle.complete()
-
 proc new*(
   T: type CodexServer,
   config: CodexConf,