Merge "Retool sqlite retries"

ironic/conf/database.py

@@ -26,9 +26,12 @@ opts = [
                 help=_('If SQLite database operation retry logic is enabled '
                        'or not. Enabled by default.')),
     cfg.IntOpt('sqlite_max_wait_for_retry',
-               default=30,
+               default=10,
                help=_('Maximum number of seconds to retry SQLite database '
-                      'locks.')),
+                      'locks, after which the original exception will be '
+                      'returned to the caller. This does not presently apply '
+                      'to internal node lock release actions and DB actions '
+                      'centered around the completion of tasks.')),
 ]
 
 

ironic/db/sqlalchemy/api.py

@@ -70,10 +70,22 @@ def wrap_sqlite_retry(f):
 
     @functools.wraps(f)
     def wrapper(*args, **kwargs):
-        if (CONF.database.sqlite_retries
+        if (not CONF.database.sqlite_retries
-                and not utils.is_ironic_using_sqlite()):
+                or not utils.is_ironic_using_sqlite()):
             return f(*args, **kwargs)
         else:
+            # NOTE(TheJulia): We likely need to see if we can separate
+            # update_node in API from the final actions of task manager
+            # actions, but that would also be an internal API change
+            # because we would likely need a special object method to
+            # call for update_node to delineate an internal save versus
+            # an external save.
+            if f.__name__ in ['update_node', 'release_node']:
+                stop = tenacity.stop_never
+            else:
+                stop = tenacity.stop_after_delay(
+                    max_delay=CONF.database.sqlite_max_wait_for_retry
+                )
             for attempt in tenacity.Retrying(
                 retry=(
                     tenacity.retry_if_exception_type(
@@ -81,13 +93,16 @@ def wrap_sqlite_retry(f):
                     & tenacity.retry_if_exception(
                         lambda e: 'database is locked' in str(e))
                 ),
-                wait=tenacity.wait_full_jitter(
+                wait=tenacity.wait_random(
-                    multiplier=0.25,
+                    min=0.1,
-                    max=CONF.database.sqlite_max_wait_for_retry),
+                    max=1,
+                ),
                 before_sleep=(
                     tenacity.before_sleep_log(LOG, logging.DEBUG)
                 ),
-                reraise=True):
+                stop=stop,
+                reraise=False,
+                retry_error_cls=exception.TemporaryFailure):
                 with attempt:
                     return f(*args, **kwargs)
     return wrapper

requirements.txt

@@ -40,7 +40,7 @@ jsonpatch!=1.20,>=1.16 # BSD
 Jinja2>=3.0.0 # BSD License (3 clause)
 keystonemiddleware>=9.5.0 # Apache-2.0
 oslo.messaging>=14.1.0 # Apache-2.0
-tenacity>=6.2.0  # Apache-2.0
+tenacity>=6.3.1  # Apache-2.0
 oslo.versionedobjects>=1.31.2 # Apache-2.0
 jsonschema>=3.2.0 # MIT
 psutil>=3.2.2 # BSD