test: Fix intermittent sync_blocks failures

1 files changed, 26 insertions, 2 deletions

diff --git a/test/functional/test_framework/test_node.py b/test/functional/test_framework/test_node.py
index e6ec3c1b2d..404c1b207b 100755
--- a/test/functional/test_framework/test_node.py
+++ b/test/functional/test_framework/test_node.py
@@ -110,7 +110,7 @@ class TestNode():
                          "--gen-suppressions=all", "--exit-on-first-error=yes",
                          "--error-exitcode=1", "--quiet"] + self.args
 
-        if self.version is None or self.version >= 190000:
+        if self.version_is_at_least(190000):
             self.args.append("-logthreadnames")
 
         self.cli = TestNodeCLI(bitcoin_cli, self.datadir)
@@ -222,6 +222,27 @@ class TestNode():
                 rpc = get_rpc_proxy(rpc_url(self.datadir, self.index, self.chain, self.rpchost), self.index, timeout=self.rpc_timeout, coveragedir=self.coverage_dir)
                 rpc.getblockcount()
                 # If the call to getblockcount() succeeds then the RPC connection is up
+                if self.version_is_at_least(190000):
+                    # getmempoolinfo.loaded is available since commit
+                    # bb8ae2c (version 0.19.0)
+                    wait_until(lambda: rpc.getmempoolinfo()['loaded'])
+                    # Wait for the node to finish reindex, block import, and
+                    # loading the mempool. Usually importing happens fast or
+                    # even "immediate" when the node is started. However, there
+                    # is no guarantee and sometimes ThreadImport might finish
+                    # later. This is going to cause intermittent test failures,
+                    # because generally the tests assume the node is fully
+                    # ready after being started.
+                    #
+                    # For example, the node will reject block messages from p2p
+                    # when it is still importing with the error "Unexpected
+                    # block message received"
+                    #
+                    # The wait is done here to make tests as robust as possible
+                    # and prevent racy tests and intermittent failures as much
+                    # as possible. Some tests might not need this, but the
+                    # overhead is trivial, and the added gurantees are worth
+                    # the minimal performance cost.
                 self.log.debug("RPC successfully started")
                 if self.use_cli:
                     return
@@ -274,6 +295,9 @@ class TestNode():
             wallet_path = "wallet/{}".format(urllib.parse.quote(wallet_name))
             return RPCOverloadWrapper(self.rpc / wallet_path, descriptors=self.descriptors)
 
+    def version_is_at_least(self, ver):
+        return self.version is None or self.version >= ver
+
     def stop_node(self, expected_stderr='', wait=0):
         """Stop the node."""
         if not self.running:
@@ -281,7 +305,7 @@ class TestNode():
         self.log.debug("Stopping node")
         try:
             # Do not use wait argument when testing older nodes, e.g. in feature_backwards_compatibility.py
-            if self.version is None or self.version >= 180000:
+            if self.version_is_at_least(180000):
                 self.stop(wait=wait)
             else:
                 self.stop()