def create_watchpoint(self, params):
"""
Create watchpoint.
Args:
- watch_condition (dict): The watch condition. The format is like:
{
"id": "tensor_too_large",
"params": [
{
"name": "abs_mean_gt",
"value": 1.1
}
]
}
- id (str): Id of condition.
- params (list[dict]): The list of param for this condition.
- watch_nodes (list[str]): The list of node names.
- watch_point_id (int): The id of watchpoint.
- search_pattern (dict): The search pattern.
- graph_name (str): The relative graph_name of the watched node.
Returns:
dict, the id of new watchpoint and metadata info.
"""
watch_condition = params.get('watch_condition')
log.info("Received create watchpoint request. WatchCondition: %s", watch_condition)
metadata_stream = self._metadata_stream
if metadata_stream.state != ServerStatus.WAITING.value:
log.error("Failed to create watchpoint as the MindSpore is not in waiting state.")
raise DebuggerCreateWatchPointError(
"Failed to create watchpoint as the MindSpore is not in waiting state.")
self._validate_watch_condition(watch_condition)
watch_nodes = self._get_watch_node_with_basic_info(
node_names=params.get('watch_nodes'),
search_pattern=params.get('search_pattern'),
graph_name=params.get('graph_name'))
validate_watch_condition(self._condition_mgr, watch_condition)
condition_id = watch_condition.get('id')
condition = self._condition_mgr.get_condition(condition_id)
condition_context = ConditionContext(metadata_stream.backend, metadata_stream.step)
if not condition.is_available(condition_context):
log.error("Failed to create watchpoint as the condition is not available.")
raise DebuggerConditionUnavailableError(
"Failed to create watchpoint as the condition is not available.")
watch_nodes = get_basic_node_info(condition.supported_target_type.value, self._graph_stream).copy()
watchpoint_stream = self._watchpoint_stream
watch_point_id = watchpoint_stream.create_watchpoint(
self._condition_mgr, watch_condition, watch_nodes, params.get('watch_point_id'))
log.info("Create watchpoint %d", watch_point_id)
metadata_stream.enable_recheck = watchpoint_stream.is_recheckable()
res = metadata_stream.get(['state', 'enable_recheck'])
res['id'] = watch_point_id
return res
def get_condition_collections(self, train_id):
"""Get default condition_collections"""
metadata_stream = self.cache_store.get_stream_handler(Streams.METADATA)
condition_context = ConditionContext(metadata_stream.backend,
metadata_stream.step)
log.debug("Train_id: %s, backend: %s", train_id,
condition_context.backend)
return self.condition_mgr.get_all_collections(condition_context)
def set_recommended_watch_points(self, set_recommended, train_id):
"""set recommended watch points."""
if not isinstance(set_recommended, bool):
log.error("Bool param should be given for set_recommended")
raise DebuggerParamValueError("Bool param should be given.")
metadata_stream = self.cache_store.get_stream_handler(Streams.METADATA)
condition_context = ConditionContext(metadata_stream.backend,
metadata_stream.step)
log.debug("Train_id: %s, backend: %s", train_id,
condition_context.backend)
res = metadata_stream.get(['state', 'enable_recheck'])
if set_recommended and not metadata_stream.recommendation_confirmed:
res['id'] = self._add_recommended_watchpoints(condition_context)
metadata_stream.recommendation_confirmed = True
return res