Harden the interop capture analysis and smppsim capture readiness (#89)

* Fail an interop run on an empty capture or a missing bind/bind_resp, and always tear down on analysis failure

* smppsim: wait for the capture sidecar to actually be capturing before node starts

* Clear the stale textdlr capture file too, and tidy the bind-check ordering

* Split the unlink comment so each line matches the unlink it explains

interop-tests/compose.smppsim.yaml

@@ -86,6 +86,13 @@ services:
       smppsim:
         condition: service_started
     command: ["dumpcap", "-i", "any", "-f", "tcp port 2775", "-w", "/captures/smppsim.pcapng"]
+    # dumpcap writes the pcapng section header as soon as it opens the interface, before any
+    # packet arrives, so a non-empty file means the capture is actually running.
+    healthcheck:
+      test: ["CMD-SHELL", "test -s /captures/smppsim.pcapng"]
+      interval: 1s
+      retries: 30
+      timeout: 2s
     volumes:
       - ./interop-tests/captures:/captures
 
@@ -101,15 +108,20 @@ services:
       smppsim-textdlr:
         condition: service_started
     command: ["dumpcap", "-i", "any", "-f", "tcp port 2775", "-w", "/captures/smppsim-textdlr.pcapng"]
+    healthcheck:
+      test: ["CMD-SHELL", "test -s /captures/smppsim-textdlr.pcapng"]
+      interval: 1s
+      retries: 30
+      timeout: 2s
     volumes:
       - ./interop-tests/captures:/captures
 
   node:
     depends_on:
       capture:
-        condition: service_started
+        condition: service_healthy
       capture-textdlr:
-        condition: service_started
+        condition: service_healthy
       smppsim:
         condition: service_healthy
       smppsim-textdlr:

interop-tests/run.py

@@ -21,6 +21,10 @@ PORT_BY_PEER = {
 	"smscsim": 2775,
 }
 
+# Every scenario in every peer's test file binds before doing anything else, so a capture missing
+# either side of that handshake never saw real traffic - the same signal as an empty capture.
+BIND_COMMANDS = ("bind_receiver", "bind_transceiver", "bind_transmitter")
+
 # The 33 SMPP commands (SMPP 3.4), by numeric command_id, for the tshark histogram.
 COMMAND_NAMES = {
 	0x00000001: "bind_receiver",
@@ -135,6 +139,9 @@ def analyse_capture(peer: str, port: int) -> tuple[int, dict[str, object]]:
 			1 for severity in collect(frame, "_ws.expert.severity") if severity == EXPERT_SEVERITY_ERROR
 		)
 
+	has_bind = any(histogram[name] for name in BIND_COMMANDS)
+	has_bind_resp = any(histogram[f"{name}_resp"] for name in BIND_COMMANDS)
+
 	print(f"frames: {len(frames)}")
 	print("commands:")
 	for name, count in sorted(histogram.items()):
@@ -142,7 +149,18 @@ def analyse_capture(peer: str, port: int) -> tuple[int, dict[str, object]]:
 	print(f"malformed: {malformed}")
 	print(f"expert errors: {expert_errors}")
 
-	status = 1 if malformed > 0 or expert_errors > 0 else 0
+	if not frames:
+		print("empty capture: no frames decoded", file=sys.stderr)
+	if not has_bind or not has_bind_resp:
+		print("no bind/bind_resp pair in capture", file=sys.stderr)
+
+	status = 1 if any((
+		not frames,
+		not has_bind,
+		not has_bind_resp,
+		malformed > 0,
+		expert_errors > 0,
+	)) else 0
 
 	return status, {
 		"commands": dict(histogram),
@@ -169,6 +187,8 @@ def main() -> int:
 	# root; overwriting one from a previous run then fails, since root here has no DAC override
 	# either - so the stale file has to go before a fresh one can be written in its place.
 	(CAPTURES_DIR / f"{args.peer}.pcapng").unlink(missing_ok=True)
+	# Also clears smppsim's textdlr capture so its healthcheck can't pass against a stale leftover.
+	(CAPTURES_DIR / f"{args.peer}-textdlr.pcapng").unlink(missing_ok=True)
 
 	tests = sh(compose_cmd(
 		args.peer, "run", "--rm", "--use-aliases", "node",
@@ -177,8 +197,9 @@ def main() -> int:
 	sh(compose_cmd(args.peer, "stop", "capture"))
 	fix_capture_ownership()
 
+	try:
 		analysis_status, _ = analyse_capture(args.peer, port)
-
+	finally:
 		if not args.keep:
 			sh(compose_cmd(args.peer, "down", "-v"))