From ffc1e226e20ec599f0f79fdc2ef878bbd1a4f04a Mon Sep 17 00:00:00 2001
From: Christian Lindig <christian.lindig@cloud.com>
Date: Fri, 8 Dec 2023 14:09:16 +0000
Subject: [PATCH] CA-386552 XSI-1534 Failed to disable pool HA after missing HA
 statefile

Backport 5a639b159f3c307a4ac017ffd3306bf161ab5947

The issue arises because:

* xapi_ha.ml, function disable_internal
* When a static VDI is removed from an SR (manually by deleting it from
  an NFS server, for example)
* As the SR is re-scanned, it is also removed from the xapi database
* But the re-scan does not remove it from the list of static VDIs in
  /etc/xensource/static-vidis
* when the function is called, it obtains the list of static VDIs (the
  UUIds)
* and looks them up in the database - which fails
* static VDIs references are further listed in Pool.ha_statefile and can
  be stale.

This patch hardens the code path to avoid failure when the statefile VDI
has been removed manually.

Signed-off-by: Christian Lindig <christian.lindig@cloud.com>
---
 ocaml/xapi/xapi_ha.ml | 29 ++++++++++++++++++++---------
 1 file changed, 20 insertions(+), 9 deletions(-)

diff --git a/ocaml/xapi/xapi_ha.ml b/ocaml/xapi/xapi_ha.ml
index eeeecf60363..d5a5f1a3b65 100644
--- a/ocaml/xapi/xapi_ha.ml
+++ b/ocaml/xapi/xapi_ha.ml
@@ -1158,6 +1158,7 @@ let emergency_ha_disable __context soft =
   )
 
 let ha_release_resources __context localhost =
+  let __FUNCTION__ = "ha_release_resources" in
   Monitor.stop () ;
 
   (* Why aren't we calling Xha_statefile.detach_existing_statefiles?
@@ -1168,14 +1169,17 @@ let ha_release_resources __context localhost =
   let statefile_vdis =
     Db.Pool.get_ha_statefiles ~__context ~self:(Helpers.get_pool ~__context)
   and deactivate_and_detach_vdi vdi_str =
-    let uuid = Db.VDI.get_uuid ~__context ~self:(Ref.of_string vdi_str) in
-    Helpers.log_exn_continue
-      (Printf.sprintf "detaching statefile VDI uuid: %s" uuid)
-      (fun () ->
-        Static_vdis.permanent_vdi_deactivate_by_uuid ~__context ~uuid ;
-        Static_vdis.permanent_vdi_detach_by_uuid ~__context ~uuid
-      )
-      ()
+    match Db.VDI.get_uuid ~__context ~self:(Ref.of_string vdi_str) with
+    | uuid ->
+        Helpers.log_exn_continue
+          (Printf.sprintf "detaching statefile VDI uuid: %s" uuid)
+          (fun () ->
+            Static_vdis.permanent_vdi_deactivate_by_uuid ~__context ~uuid ;
+            Static_vdis.permanent_vdi_detach_by_uuid ~__context ~uuid
+          )
+          ()
+    | exception _e ->
+        warn "%s: VDI %s not found in database" __FUNCTION__ vdi_str
   in
   List.iter deactivate_and_detach_vdi statefile_vdis ;
   (* Deactivate and detach any metadata VDIs *)
@@ -1516,9 +1520,16 @@ let abort_new_master ~__context ~address =
 let disable_internal __context =
   debug "Disabling HA on the Pool" ;
   let pool = Helpers.get_pool ~__context in
+  (* Avoid stale static VDIs *)
+  Static_vdis.gc () ;
   (* Find the HA metadata and statefile VDIs for later *)
   let statefile_vdis =
-    List.map Ref.of_string (Db.Pool.get_ha_statefiles ~__context ~self:pool)
+    let is_valid ref =
+      if Db.is_valid_ref __context ref then Some ref else None
+    in
+    Db.Pool.get_ha_statefiles ~__context ~self:pool
+    |> List.map Ref.of_string
+    |> List.filter_map is_valid
   in
   let metadata_vdis =
     List.map