sambacc: add a retry loop to ctdb.monitor_cluster_meta_changes

Add a loop that tries the `ctdb reloadnodes` command after an increasing
delay. This is an attempt to fix a condition where ctdbd is apparently
not ready to handle the `ctdb reloadnodes` command. In this case the
command would be run, but fail and an exception would be raised in the
monitor_cluster_meta_changes function would raise an exception. This
would be caught by the command-level retry loop. However, this
command-level retry loop will simply re-run monitor_cluster_meta_changes
and this function now no longer has the same initial clustermeta state
and has effectively "forgotten" that it needs to run reloadnodes.  This
new retry loop adds a level of error handling inside the
monitor_cluster_meta_changes function so that we will retry with a
bounded number of attempts.

Signed-off-by: John Mulligan <[email protected]>

Author: phlogistonjohn

sambacc/ctdb.py

@@ -20,6 +20,7 @@
 import logging
 import os
 import subprocess
+import time
 import typing
 
 from sambacc import config
@@ -572,7 +573,23 @@ def monitor_cluster_meta_changes(
         if nodes_file_path:
             _logger.info("updating nodes file: %s", nodes_file_path)
             _save_nodes(nodes_file_path, expected_nodes)
-        _maybe_reload_nodes(leader_locator, reload_all=reload_all)
+        _maybe_reload_nodes_retry(leader_locator, reload_all=reload_all)
+
+
+def _maybe_reload_nodes_retry(
+    leader_locator: typing.Optional[leader.LeaderLocator] = None,
+    reload_all: bool = False,
+    *,
+    tries: int = 5,
+) -> None:
+    for idx in range(tries):
+        time.sleep(1 << idx)
+        try:
+            _maybe_reload_nodes(leader_locator, reload_all=reload_all)
+            return
+        except subprocess.CalledProcessError:
+            _logger.exception("failed to execute reload nodes command")
+    raise RuntimeError("exceeded retries running reload nodes command")
 
 
 def _maybe_reload_nodes(