From ac2255bdc44157af456a4f944e2c653cf715160c Mon Sep 17 00:00:00 2001
From: Christian Lindig <christian.lindig@cloud.com>
Date: Thu, 19 Dec 2024 14:55:06 +0000
Subject: [PATCH] XSI-1773 improve logging if service file unexpectedly exists

We have seen failures where a service file unexpectedly exists. It could
have been left behind but a failed stop but we don't have evidence for
that. To help with this, provide more details of the file found.

Signed-off-by: Christian Lindig <christian.lindig@cloud.com>
---
 ocaml/forkexecd/lib/dune            |  1 +
 ocaml/forkexecd/lib/fe_systemctl.ml | 24 +++++++++++++++++++-----
 2 files changed, 20 insertions(+), 5 deletions(-)

diff --git a/ocaml/forkexecd/lib/dune b/ocaml/forkexecd/lib/dune
index 749f173b977..e8dd8c8312e 100644
--- a/ocaml/forkexecd/lib/dune
+++ b/ocaml/forkexecd/lib/dune
@@ -13,6 +13,7 @@
    xapi-log
    xapi-stdext-pervasives
    xapi-stdext-unix
+   xapi-stdext-date
    xapi-tracing
  )
  (preprocess (per_module ((pps ppx_deriving_rpc) Fe))))
diff --git a/ocaml/forkexecd/lib/fe_systemctl.ml b/ocaml/forkexecd/lib/fe_systemctl.ml
index b36ee6674ae..046396002ca 100644
--- a/ocaml/forkexecd/lib/fe_systemctl.ml
+++ b/ocaml/forkexecd/lib/fe_systemctl.ml
@@ -130,14 +130,28 @@ let is_active ~service =
   in
   Unix.WEXITED 0 = status
 
-let exists ~service =
-  Sys.file_exists (Filename.concat run_path (service ^ ".service"))
+(** path to service file *)
+let path service = Filename.concat run_path (service ^ ".service")
+
+(** does [service] file exist *)
+let exists ~service = Sys.file_exists (path service)
+
+(** creation time of [path] as a string *)
+let ctime path =
+  let ctime = Unix.((stat path).st_ctime) in
+  Xapi_stdext_date.Date.(of_unix_time ctime |> to_rfc3339)
 
 let start_transient ?env ?properties ?(exec_ty = Type.Simple) ~service cmd args
     =
-  if exists ~service then
-    (* this can only happen if there is a bug in the caller *)
-    invalid_arg (Printf.sprintf "Tried to start %s twice" service) ;
+  ( match exists ~service with
+  | true ->
+      (* this can only happen if there is a bug in the caller *)
+      let path = path service in
+      let invalid fmt = Printf.ksprintf invalid_arg fmt in
+      invalid "Tried to start %s twice: %s exists (%s)" service path (ctime path)
+  | false ->
+      ()
+  ) ;
   try start_transient ?env ?properties ~exec_ty ~service cmd args
   with e ->
     Backtrace.is_important e ;