Function build_exec_summary_table

shell/impala_shell/exec_summary.py:25–177

Direct translation of Coordinator::PrintExecSummary() to recursively build a list of rows of summary statistics, one per exec node summary: the TExecSummary object that contains all the summary data idx: the index of the node to print indent_level: the number of spaces to print before wri

(summary, idx, indent_level, new_indent_level, output,
                             is_prettyprint=True, separate_prefix_column=False)

def build_exec_summary_table(summary, idx, indent_level, new_indent_level, output,
                             is_prettyprint=True, separate_prefix_column=False):
  """Direct translation of Coordinator::PrintExecSummary() to recursively build a list
  of rows of summary statistics, one per exec node

  summary: the TExecSummary object that contains all the summary data

  idx: the index of the node to print

  indent_level: the number of spaces to print before writing the node's label, to give
  the appearance of a tree. The 0th child of a node has the same indent_level as its
  parent. All other children have an indent_level of one greater than their parent.

  new_indent_level: If true, this indent level is different from the previous row's.

  output: the list of rows into which to append the rows produced for this node and its
  children.

  is_prettyprint: Optional. If True, print time, units, and bytes columns in pretty
  printed format.

  separate_prefix_column: Optional. If True, the prefix and operator name will be
  returned as separate column. Otherwise, prefix and operater name will be concatenated
  into single column.

  Returns the index of the next exec node in summary.exec_nodes that should be
  processed, used internally to this method only.
  """
  if not summary.nodes:
    # Summary nodes is empty or None. Nothing to build.
    return
  assert idx < len(summary.nodes), (
      "Index ({0}) must be less than exec summary count ({1})").format(
          idx, len(summary.nodes))

  attrs = ["latency_ns", "cpu_time_ns", "cardinality", "memory_used"]

  # Initialise aggregate and maximum stats
  agg_stats, max_stats = TExecStats(), TExecStats()
  for attr in attrs:
    setattr(agg_stats, attr, 0)
    setattr(max_stats, attr, 0)

  node = summary.nodes[idx]
  instances = 0
  if node.exec_stats:
    # exec_stats is not None or an empty list.
    instances = len(node.exec_stats)
    for stats in node.exec_stats:
      for attr in attrs:
        val = getattr(stats, attr)
        if val is not None:
          setattr(agg_stats, attr, getattr(agg_stats, attr) + val)
          setattr(max_stats, attr, max(getattr(max_stats, attr), val))
    avg_time = agg_stats.latency_ns / instances
  else:
    avg_time = 0