tests/default.nix: Work around a race-condition in the acme cert waiting

1 files changed, 21 insertions, 5 deletions

diff --git a/tests/default.nix b/tests/default.nix
index 48fa9ec..9bab516 100644
--- a/tests/default.nix
+++ b/tests/default.nix
@@ -136,21 +136,37 @@
               return status == 0
 
             wasAlive: bool
+            deadCounter: int
             def run_and_wait(host, unit: str):
               global wasAlive
+              global deadCounter
+
               wasAlive = False
+              deadCounter = 0
+
               def wait_state(_last_try: bool) -> bool:
                 global wasAlive
+                global deadCounter
 
                 state = host.get_unit_property(unit, "SubState")
-                if state == "failed":
-                    assert False, f'Unit "{unit}" reached state "{state}"'
+                host.log(f"(waiting) {unit} -> {state}")
 
-                if state == "start" or state == "active" or state == "running":
+                match state:
+                  case "failed":
+                    assert False, f'Unit "{unit}" reached state "{state}"'
+                  case "start" | "active" | "running":
                     wasAlive = True
+                  case "dead":
+                    if wasAlive:
+                      return True
+                    else:
+                      if deadCounter >= 5:
+                        return True
 
-                host.log(f"(waiting) {unit} -> {state}")
-                return wasAlive and state == "dead"
+                      deadCounter = deadCounter + 1
+                      return False
+
+                return False
 
               with host.nested(f"Waiting for unit {unit}"):
                 retry(wait_state)