impala/testdata/bin/generate-schema-statements.py

#!/usr/bin/env impala-python
# Copyright (c) 2012 Cloudera, Inc. All rights reserved.

# This script generates the "CREATE TABLE", "INSERT", and "LOAD" statements for loading
# test data and writes them to create-*-generated.sql and
# load-*-generated.sql. These files are then executed by hive or impala, depending
# on their contents. Additionally, for hbase, the file is of the form
# create-*hbase*-generated.create.
#
# The statements that are generated are based on an input test vector
# (read from a file) that describes the coverage desired. For example, currently
# we want to run benchmarks with different data sets, across different file types, and
# with different compression algorithms set. To improve data loading performance this
# script will generate an INSERT INTO statement to generate the data if the file does
# not already exist in HDFS. If the file does already exist in HDFS then we simply issue a
# LOAD statement which is much faster.
#
# The input test vectors are generated via the generate_test_vectors.py so
# ensure that script has been run (or the test vector files already exist) before
# running this script.
#
# Note: This statement generation is assuming the following data loading workflow:
# 1) Load all the data in the specified source table
# 2) Create tables for the new file formats and compression types
# 3) Run INSERT OVERWRITE TABLE SELECT * from the source table into the new tables
#    or LOAD directly if the file already exists in HDFS.
import collections
import csv
import glob
import json
import math
import os
import random
import re
import shutil
import subprocess
import sys
import tempfile
from itertools import product
from optparse import OptionParser
from tests.util.test_file_parser import *
from tests.common.test_dimensions import *

parser = OptionParser()
parser.add_option("-e", "--exploration_strategy", dest="exploration_strategy",
                  default="core", help="The exploration strategy for schema gen: 'core',"\
                  " 'pairwise', or 'exhaustive'")
parser.add_option("--hive_warehouse_dir", dest="hive_warehouse_dir",
                  default="/test-warehouse",
                  help="The HDFS path to the base Hive test warehouse directory")
parser.add_option("-w", "--workload", dest="workload",
                  help="The workload to generate schema for: tpch, hive-benchmark, ...")
parser.add_option("-s", "--scale_factor", dest="scale_factor", default="",
                  help="An optional scale factor to generate the schema for")
parser.add_option("-f", "--force_reload", dest="force_reload", action="store_true",
                  default= False, help='Skips HDFS exists check and reloads all tables')
parser.add_option("-v", "--verbose", dest="verbose", action="store_true",
                  default = False, help="If set, outputs additional logging.")
parser.add_option("-b", "--backend", dest="backend", default="localhost:21000",
                  help="Backend connection to use, default: localhost:21000")
parser.add_option("--table_names", dest="table_names", default=None,
                  help="Only load the specified tables - specified as a comma-seperated "\
                  "list of base table names")
parser.add_option("--table_formats", dest="table_formats", default=None,
                  help="Override the test vectors and load using the specified table "\
                  "formats. Ex. --table_formats=seq/snap/block,text/none")
parser.add_option("--hdfs_namenode", dest="hdfs_namenode", default="localhost:20500",
                  help="HDFS name node for Avro schema URLs, default localhost:20500")
(options, args) = parser.parse_args()

if options.workload is None:
  print "A workload name must be specified."
  parser.print_help()
  sys.exit(1)

WORKLOAD_DIR = os.path.join(os.environ['IMPALA_HOME'], 'testdata', 'workloads')
DATASET_DIR = os.path.join(os.environ['IMPALA_HOME'], 'testdata', 'datasets')
SQL_OUTPUT_DIR = os.environ['IMPALA_DATA_LOADING_SQL_DIR']
AVRO_SCHEMA_DIR = "avro_schemas"
IMPALA_SUPPORTED_INSERT_FORMATS = ['parquet', 'hbase', 'text', 'kudu']

COMPRESSION_TYPE = "SET mapred.output.compression.type=%s;"
COMPRESSION_ENABLED = "SET hive.exec.compress.output=%s;"
COMPRESSION_CODEC = "SET mapred.output.compression.codec=%s;"
AVRO_COMPRESSION_CODEC = "SET avro.output.codec=%s;"
SET_DYNAMIC_PARTITION_STATEMENT = "SET hive.exec.dynamic.partition=true;"
SET_PARTITION_MODE_NONSTRICT_STATEMENT = "SET hive.exec.dynamic.partition.mode=nonstrict;"
SET_HIVE_INPUT_FORMAT = "SET mapred.max.split.size=256000000;\n"\
                        "SET hive.input.format=org.apache.hadoop.hive.ql.io.%s;\n"
SET_HIVE_HBASE_BULK_LOAD = "SET hive.hbase.bulk = true"
FILE_FORMAT_IDX = 0
DATASET_IDX = 1
CODEC_IDX = 2
COMPRESSION_TYPE_IDX = 3

COMPRESSION_MAP = {'def': 'org.apache.hadoop.io.compress.DefaultCodec',
                   'gzip': 'org.apache.hadoop.io.compress.GzipCodec',
                   'bzip': 'org.apache.hadoop.io.compress.BZip2Codec',
                   'snap': 'org.apache.hadoop.io.compress.SnappyCodec',
                   'lzo': 'com.hadoop.compression.lzo.LzopCodec',
                   'none': ''
                  }

AVRO_COMPRESSION_MAP = {
  'def': 'deflate',
  'snap': 'snappy',
  'none': '',
  }

FILE_FORMAT_MAP = {
  'text': 'TEXTFILE',
  'seq': 'SEQUENCEFILE',
  'rc': 'RCFILE',
  'parquet': 'PARQUET',
  'text_lzo':
    "\nINPUTFORMAT 'com.hadoop.mapred.DeprecatedLzoTextInputFormat'" +
    "\nOUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.HiveIgnoreKeyTextOutputFormat'",
  'avro': 'AVRO',
  'hbase': "'org.apache.hadoop.hive.hbase.HBaseStorageHandler'",
  'kudu': "'com.cloudera.kudu.hive.KuduStorageHandler'",
  }

HIVE_TO_AVRO_TYPE_MAP = {
  'STRING': 'string',
  'INT': 'int',
  'TINYINT': 'int',
  'SMALLINT': 'int',
  'BIGINT': 'long',
  'BOOLEAN': 'boolean',
  'FLOAT': 'float',
  'DOUBLE': 'double',
  # Avro has no timestamp type, so convert to string
  # TODO: this allows us to create our Avro test tables, but any tests that use
  # a timestamp column will fail. We probably want to convert back to timestamps
  # in our tests.
  'TIMESTAMP': 'string',
  }

PARQUET_ALTER_STATEMENT = "ALTER TABLE %(table_name)s SET\n\
     SERDEPROPERTIES ('blocksize' = '1073741824', 'compression' = '%(compression)s');"

HBASE_CREATE_STATEMENT = """
CREATE EXTERNAL TABLE IF NOT EXISTS {{db_name}}{{db_suffix}}.{{table_name}} (
{columns})
STORED BY {{file_format}}
WITH SERDEPROPERTIES (
  "hbase.columns.mapping" =
  "{hbase_column_mapping}")
{tbl_properties}{{hdfs_location}}"""

KNOWN_EXPLORATION_STRATEGIES = ['core', 'pairwise', 'exhaustive', 'lzo']

def build_create_statement(table_template, table_name, db_name, db_suffix,
                           file_format, compression, hdfs_location):
  create_stmt = 'CREATE DATABASE IF NOT EXISTS %s%s;\n' % (db_name, db_suffix)
  if (options.force_reload):
    create_stmt += 'DROP TABLE IF EXISTS %s%s.%s;\n' % (db_name, db_suffix, table_name)
  if compression == 'lzo':
    file_format = '%s_%s' % (file_format, compression)
  # hbase / kudu tables are external, and not read from hdfs. We don't need an
  # hdfs_location.
  if file_format in ['hbase', 'kudu']:
    hdfs_location = str()
    # Remove location part from the format string
    table_template = table_template.replace("LOCATION '{hdfs_location}'", "")

  create_stmt += table_template.format(db_name=db_name,
                                       db_suffix=db_suffix,
                                       table_name=table_name,
                                       file_format=FILE_FORMAT_MAP[file_format],
                                       hdfs_location=hdfs_location)
  return create_stmt

def build_table_template(file_format, columns, partition_columns, row_format,
                         avro_schema_dir, table_name, table_properties):
  if file_format == 'hbase':
    return build_hbase_create_stmt_in_hive(columns, partition_columns, table_name)

  partitioned_by = str()
  if partition_columns:
    partitioned_by = 'PARTITIONED BY (%s)' % ', '.join(partition_columns.split('\n'))

  row_format_stmt = str()
  if row_format:
    row_format_stmt = 'ROW FORMAT ' + row_format

  file_format_string = str()
  if file_format != 'kudu':
    file_format_string = "STORED AS {file_format}"

  tblproperties_clause = "TBLPROPERTIES (\n{0}\n)"
  tblproperties = {}

  external = "EXTERNAL"

  if file_format == 'avro':
    tblproperties["avro.schema.url"] = "hdfs://%s/%s/%s/{table_name}.json" \
      % (options.hdfs_namenode, options.hive_warehouse_dir, avro_schema_dir)
  elif file_format == 'parquet':
    row_format_stmt = str()
  elif file_format == 'kudu':
    # Fetch KUDU host and port from environment
    kudu_master = os.getenv("KUDU_MASTER_ADDRESS", "127.0.0.1")
    kudu_master_port = os.getenv("KUDU_MASTER_PORT", "7051")
    row_format_stmt = str()
    tblproperties["storage_handler"] = "com.cloudera.kudu.hive.KuduStorageHandler"
    tblproperties["kudu.master_addresses"] = \
      "{0}:{1}".format(kudu_master, kudu_master_port)
    tblproperties["kudu.table_name"] = table_name
    tblproperties["kudu.key_columns"] = columns.split("\n")[0].split(" ")[0]
    # Kudu's test tables are managed.
    external = ""

  # Read the properties specified in the TABLE_PROPERTIES section. When the specified
  # properties have the same key as a default property, the value for the specified
  # property is used.
  if table_properties:
    for table_property in table_properties.split("\n"):
      format_prop = table_property.split(":")
      if format_prop[0] == file_format:
        key_val = format_prop[1].split("=");
        tblproperties[key_val[0]] = key_val[1]

  all_tblproperties = []
  for key, value in tblproperties.iteritems():
    all_tblproperties.append("'{0}' = '{1}'".format(key, value))

  # If there are no properties to set avoid the TBLPROPERTIES clause altogether.
  if not all_tblproperties:
    tblproperties_clause = ""
  else:
    tblproperties_clause = tblproperties_clause.format(",\n".join(all_tblproperties))

  # Note: columns are ignored but allowed if a custom serde is specified
  # (e.g. Avro)
  stmt = """
CREATE {external} TABLE IF NOT EXISTS {{db_name}}{{db_suffix}}.{{table_name}} (
{columns})
{partitioned_by}
{row_format}
{file_format_string}
LOCATION '{{hdfs_location}}'
{tblproperties}
""".format(
    external=external,
    row_format=row_format_stmt,
    columns=',\n'.join(columns.split('\n')),
    partitioned_by=partitioned_by,
    tblproperties=tblproperties_clause,
    file_format_string=file_format_string
    ).strip()

  # Remove empty lines from the stmt string.  There is an empty line for
  # each of the sections that didn't have anything (e.g. partitioned_by)
  stmt = os.linesep.join([s for s in stmt.splitlines() if s])
  stmt += ';'
  return stmt

def build_hbase_create_stmt_in_hive(columns, partition_columns, table_name):
  # The hbase create statement differs sufficiently from the generic create to justify a
  # separate method. Specifically, STORED AS becomes STORED BY. There is section called
  # serdeproperties, the partition colmns have to be appended to columns in the schema.
  columns = columns.split('\n')
  # partition columns have to be appended to the columns in the schema.
  # PARTITIONED BY is not supported and does not make sense for HBase.
  if partition_columns:
    columns.extend(partition_columns.split('\n'))
  # stringid is a special case. It still points to functional_hbase.alltypesagg
  if 'stringid' not in table_name:
    tbl_properties = ('TBLPROPERTIES("hbase.table.name" = '
                      '"{db_name}{db_suffix}.{table_name}")')
  else:
    tbl_properties = ('TBLPROPERTIES("hbase.table.name" = '
                      '"{db_name}{db_suffix}.alltypesagg")')
  # build hbase column mapping, the first column is implicitly the primary key
  # which has a diffrerent representation [:key]
  hbase_column_mapping = ["d:%s" % c.split(' ')[0] for c in columns[1:]]
  hbase_column_mapping = ":key," + ','.join(hbase_column_mapping)
  stmt = HBASE_CREATE_STATEMENT.format(
    columns=',\n'.join(columns),
    hbase_column_mapping=hbase_column_mapping,
    tbl_properties=tbl_properties,
    ).strip()
  return stmt + ';'

def avro_schema(columns):
  record = {
    "name": "a", # doesn't matter
    "type": "record",
    "fields": list()
    }
  for column_spec in columns.strip().split('\n'):
    # column_spec looks something like "col_name col_type COMMENT comment"
    # (comment may be omitted, we don't use it)
    name = column_spec.split()[0]

    if "DECIMAL" in column_spec.upper():
      if column_spec.split()[1].upper() == "DECIMAL":
        # No scale and precision specified, use defaults
        scale = 0
        precision = 9
      else:
        # Parse out scale and precision from decimal type
        m = re.search("DECIMAL\((?P<precision>.*),(?P<scale>.*)\)", column_spec.upper())
        assert m, "Could not parse decimal column spec: " + column_spec
        scale = int(m.group('scale'))
        precision = int(m.group('precision'))
      type = {"type": "bytes", "logicalType": "decimal", "precision": precision,
              "scale": scale}
    else:
      hive_type = column_spec.split()[1]
      type = HIVE_TO_AVRO_TYPE_MAP[hive_type.upper()]

    record["fields"].append(
      {'name': name,
       'type': [type, "null"]}) # all columns nullable

  return json.dumps(record)

def build_compression_codec_statement(codec, compression_type, file_format):
  codec = AVRO_COMPRESSION_MAP[codec] if file_format == 'avro' else COMPRESSION_MAP[codec]
  if not codec:
    return str()
  return (AVRO_COMPRESSION_CODEC % codec) if file_format == 'avro' else (
    COMPRESSION_TYPE % compression_type.upper() + '\n' + COMPRESSION_CODEC % codec)

def build_codec_enabled_statement(codec):
  compression_enabled = 'false' if codec == 'none' else 'true'
  return COMPRESSION_ENABLED % compression_enabled

def build_insert_into_statement(insert, db_name, db_suffix, table_name, file_format,
                                hdfs_path, for_impala=False):
  insert_statement = insert.format(db_name=db_name,
                                   db_suffix=db_suffix,
                                   table_name=table_name,
                                   hdfs_location=hdfs_path)

  # Kudu tables are managed and don't support OVERWRITE, so we replace OVERWRITE
  # with INTO to make this a regular INSERT.
  if file_format == 'kudu':
    insert_statement = insert_statement.replace("OVERWRITE", "INTO")

  if for_impala:
    return insert_statement

  statement = SET_PARTITION_MODE_NONSTRICT_STATEMENT + "\n"
  statement += SET_DYNAMIC_PARTITION_STATEMENT + "\n"
  statement += "set hive.auto.convert.join=true;\n"

  # For some reason (hive bug?) we need to have the CombineHiveInputFormat set
  # for cases where we are compressing in bzip or lzo on certain tables that
  # have multiple files.
  if 'multi' in table_name and ('bzip' in db_suffix or 'lzo' in db_suffix):
    statement += SET_HIVE_INPUT_FORMAT % "CombineHiveInputFormat"
  else:
    statement += SET_HIVE_INPUT_FORMAT % "HiveInputFormat"
  return statement + insert_statement

def build_hbase_insert(db_name, db_suffix, table_name):
  hbase_insert = SET_HIVE_HBASE_BULK_LOAD + ';\n'
  hbase_insert += ("INSERT OVERWRITE TABLE {db_name}{db_suffix}.{table_name}"
                   " SELECT * FROM {db_name}.{table_name};\n").\
                   format(db_name=db_name, db_suffix=db_suffix,table_name=table_name)
  return hbase_insert

def build_insert(insert, db_name, db_suffix, file_format,
                 codec, compression_type, table_name, hdfs_path, create_hive=False):
  # HBASE inserts don't need the hive options to be set, and don't require and HDFS
  # file location, so they're handled separately.
  if file_format == 'hbase' and not create_hive:
    return build_hbase_insert(db_name, db_suffix, table_name)
  output = build_codec_enabled_statement(codec) + "\n"
  output += build_compression_codec_statement(codec, compression_type, file_format) + "\n"
  output += build_insert_into_statement(insert, db_name, db_suffix,
                                        table_name, file_format, hdfs_path) + "\n"
  return output

def build_load_statement(load_template, db_name, db_suffix, table_name):
  # hbase does not need the hdfs path.
  if table_name.startswith('hbase'):
    load_template = load_template.format(table_name=table_name,
                                         db_name=db_name,
                                         db_suffix=db_suffix)
  else:
    load_template = load_template.format(table_name=table_name,
                                         db_name=db_name,
                                         db_suffix=db_suffix,
                                         impala_home = os.environ['IMPALA_HOME'])
  return load_template

def build_hbase_create_stmt(db_name, table_name, column_families):
  hbase_table_name = "{db_name}_hbase.{table_name}".format(db_name=db_name,
                                                           table_name=table_name)
  create_stmt = list()
  create_stmt.append("disable '%s'" % hbase_table_name)
  create_stmt.append("drop '%s'" % hbase_table_name)
  column_families = ','.join(["'{0}'".format(cf) for cf in column_families.splitlines()])
  create_stmt.append("create '%s', %s" % (hbase_table_name, column_families))
  return create_stmt

def build_db_suffix(file_format, codec, compression_type):
  if file_format == 'text' and codec == 'none':
    return ''
  elif codec == 'none':
    return '_%s' % (file_format)
  elif compression_type == 'record':
    return '_%s_record_%s' % (file_format, codec)
  else:
    return '_%s_%s' % (file_format, codec)

# Does a hdfs directory listing and returns array with all the subdir names.
def get_hdfs_subdirs_with_data(path):
  tmp_file = tempfile.TemporaryFile("w+")
  cmd = "hadoop fs -du %s | grep -v '^0' | awk '{print $3}'" % path
  subprocess.call([cmd], shell = True, stderr = open('/dev/null'), stdout = tmp_file)
  tmp_file.seek(0)

  # Results look like:
  # <acls> -  <user> <group> <date> /directory/subdirectory
  # So to get subdirectory names just return everything after the last '/'
  return [line[line.rfind('/') + 1:].strip() for line in tmp_file.readlines()]

class Statements(object):
  """Simple container object for storing SQL statements to be output to a
  file. Useful for ordering the statements correctly."""
  def __init__(self):
    self.create = list()
    self.load = list()
    self.load_base = list()

  def write_to_file(self, filename):
    # If there is no content to write, skip
    if self.__is_empty(): return
    output = self.create + self.load_base + self.load
    with open(filename, 'w') as f:
      f.write('\n\n'.join(output))

  def __is_empty(self):
    return not (self.create or self.load or self.load_base)

def eval_section(section_str):
  """section_str should be the contents of a section (i.e. a string). If section_str
  starts with `, evaluates section_str as a shell command and returns the
  output. Otherwise returns section_str."""
  if not section_str.startswith('`'): return section_str
  cmd = section_str[1:]
  # Use bash explicitly instead of setting shell=True so we get more advanced shell
  # features (e.g. "for i in {1..n}")
  p = subprocess.Popen(['/bin/bash', '-c', cmd], stdout=subprocess.PIPE)
  stdout, stderr = p.communicate()
  if stderr: print stderr
  assert p.returncode == 0
  return stdout.strip()

def generate_statements(output_name, test_vectors, sections,
                        schema_include_constraints, schema_exclude_constraints,
                        schema_only_constraints):
  # TODO: This method has become very unwieldy. It has to be re-factored sooner than
  # later.
  # Parquet statements to be executed separately by Impala
  hive_output = Statements()
  hbase_output = Statements()
  hbase_post_load = Statements()

  table_names = None
  if options.table_names:
    table_names = [name.lower() for name in options.table_names.split(',')]
  existing_tables = get_hdfs_subdirs_with_data(options.hive_warehouse_dir)
  for row in test_vectors:
    impala_output = Statements()
    impala_load = Statements()
    file_format, data_set, codec, compression_type =\
        [row.file_format, row.dataset, row.compression_codec, row.compression_type]
    table_format = '%s/%s/%s' % (file_format, codec, compression_type)
    for section in sections:
      table_name = section['BASE_TABLE_NAME']
      db_suffix = build_db_suffix(file_format, codec, compression_type)
      db_name = '{0}{1}'.format(data_set, options.scale_factor)
      db = '{0}{1}'.format(db_name, db_suffix)


      if table_names and (table_name.lower() not in table_names):
        print 'Skipping table: %s.%s, table is not in specified table list' % (db, table_name)
        continue

      if table_format in schema_only_constraints and \
         table_name.lower() not in schema_only_constraints[table_format]:
        print ('Skipping table: %s.%s, \'only\' constraint for format did not '
               'include this table.') % (db, table_name)
        continue

      if schema_include_constraints[table_name.lower()] and \
         table_format not in schema_include_constraints[table_name.lower()]:
        print 'Skipping \'%s.%s\' due to include constraint match.' % (db, table_name)
        continue

      if schema_exclude_constraints[table_name.lower()] and\
         table_format in schema_exclude_constraints[table_name.lower()]:
        print 'Skipping \'%s.%s\' due to exclude constraint match.' % (db, table_name)
        continue

      alter = section.get('ALTER')
      create = section['CREATE']
      create_hive = section['CREATE_HIVE']

      if file_format == 'kudu':
        create_kudu = section["CREATE_KUDU"]
      else:
        create_kudu = None

      table_properties = section['TABLE_PROPERTIES']
      insert = eval_section(section['DEPENDENT_LOAD'])
      load = eval_section(section['LOAD'])
      # For some datasets we may want to use a different load strategy when running local
      # tests versus tests against large scale factors. The most common reason is to
      # reduce he number of partitions for the local test environment
      if not options.scale_factor and section['LOAD_LOCAL']:
        load = section['LOAD_LOCAL']

      columns = eval_section(section['COLUMNS'])
      partition_columns = section['PARTITION_COLUMNS']
      row_format = section['ROW_FORMAT']

      # Force reloading of the table if the user specified the --force option or
      # if the table is partitioned and there was no ALTER section specified. This is to
      # ensure the partition metadata is always properly created. The ALTER section is
      # used to create partitions, so if that section exists there is no need to force
      # reload.
      # TODO: Rename the ALTER section to ALTER_TABLE_ADD_PARTITION
      force_reload = options.force_reload or (partition_columns and not alter)

      hdfs_location = '{0}.{1}{2}'.format(db_name, table_name, db_suffix)
      # hdfs file names for hive-benchmark and functional datasets are stored
      # directly under /test-warehouse
      # TODO: We should not need to specify the hdfs file path in the schema file.
      # This needs to be done programmatically.
      if data_set in ['hive-benchmark', 'functional']:
        hdfs_location = hdfs_location.split('.')[-1]
      # hive does not allow hyphenated table names.
      if data_set == 'hive-benchmark':
        db_name = '{0}{1}'.format('hivebenchmark', options.scale_factor)
      data_path = os.path.join(options.hive_warehouse_dir, hdfs_location)

      # Empty tables (tables with no "LOAD" sections) are assumed to be used for insert
      # testing. Since Impala currently only supports inserting into TEXT, PARQUET and
      # HBASE we need to create these tables with a supported insert format.
      create_file_format = file_format
      create_codec = codec
      if not (section['LOAD'] or section['LOAD_LOCAL'] or section['DEPENDENT_LOAD']):
        create_codec = 'none'
        create_file_format = file_format
        if file_format not in IMPALA_SUPPORTED_INSERT_FORMATS:
          create_file_format = 'text'

      output = impala_output
      if create_hive or file_format == 'hbase':
        output = hive_output
      elif codec == 'lzo':
        # Impala CREATE TABLE doesn't allow INPUTFORMAT.
        output = hive_output

      # TODO: Currently, Kudu does not support partitioned tables via Impala
      if file_format == 'kudu' and partition_columns != '':
        print "Ignore partitions on Kudu"
        continue

      # If a CREATE section is provided, use that. Otherwise a COLUMNS section
      # must be provided (and optionally PARTITION_COLUMNS and ROW_FORMAT
      # sections), which is used to generate the create table statement.
      if create_hive:
        table_template = create_hive
      elif create_kudu:
        table_template = create_kudu
      elif create:
        table_template = create
        if file_format in ['avro', 'hbase', 'kudu']:
          # We don't know how to generalize CREATE sections to Avro and hbase.
          print ("CREATE section not supported with %s, "
                 "skipping: '%s'" % (file_format, table_name))
          continue
      elif columns:
        avro_schema_dir = "%s/%s" % (AVRO_SCHEMA_DIR, data_set)
        table_template = build_table_template(
          create_file_format, columns, partition_columns,
          row_format, avro_schema_dir, table_name, table_properties)
        # Write Avro schema to local file
        if file_format == 'avro':
          if not os.path.exists(avro_schema_dir):
            os.makedirs(avro_schema_dir)
          with open("%s/%s.json" % (avro_schema_dir, table_name),"w") as f:
            f.write(avro_schema(columns))
      else:
        table_template = None

      if table_template:
        output.create.append(build_create_statement(table_template, table_name, db_name,
            db_suffix, create_file_format, create_codec, data_path))
      # HBASE create table
      if file_format == 'hbase':
        # If the HBASE_COLUMN_FAMILIES section does not exist, default to 'd'
        column_families = section.get('HBASE_COLUMN_FAMILIES', 'd')
        hbase_output.create.extend(build_hbase_create_stmt(db_name, table_name,
            column_families))
        hbase_post_load.load.append("flush '%s_hbase.%s'\n" % (db_name, table_name))

      # The ALTER statement in hive does not accept fully qualified table names so
      # insert a use statement. The ALTER statement is skipped for HBASE as it's
      # used for adding partitions.
      # TODO: Consider splitting the ALTER subsection into specific components. At the
      # moment, it assumes we're only using ALTER for partitioning the table.
      if alter and file_format not in ("hbase", "kudu"):
        use_db = 'USE {db_name};\n'.format(db_name=db)
        if output == hive_output and codec == 'lzo':
          # Hive ALTER TABLE ADD PARTITION doesn't handle null partitions, so
          # we can't run the ALTER section in this case.
          if options.force_reload:
            # IMPALA-2278: Hive INSERT OVERWRITE won't clear out partition directories
            # that weren't already added to the table. So, for force reload, manually
            # delete the partition directories.
            output.create.append(("DFS -rm -R {data_path};").format(
              data_path=data_path));
          else:
            # If this is not a force reload use msck repair to add the partitions
            # into the table.
            output.create.append(use_db + 'msck repair table %s;' % (table_name))
        else:
          output.create.append(use_db + alter.format(table_name=table_name))

      # If the directory already exists in HDFS, assume that data files already exist
      # and skip loading the data. Otherwise, the data is generated using either an
      # INSERT INTO statement or a LOAD statement.
      if not force_reload and hdfs_location in existing_tables:
        print 'HDFS path:', data_path, 'contains data. Data loading can be skipped.'
      else:
        print 'HDFS path:', data_path, 'does not exists or is empty. Data will be loaded.'
        if not db_suffix:
          if load:
            hive_output.load_base.append(build_load_statement(load, db_name,
                                                              db_suffix, table_name))
          else:
            print 'Empty base table load for %s. Skipping load generation' % table_name
        elif file_format in ['kudu', 'parquet']:
          if insert:
            impala_load.load.append(build_insert_into_statement(insert, db_name,
                db_suffix, table_name, file_format, data_path, for_impala=True))
          else:
            print 'Empty parquet/kudu load for table %s. Skipping insert generation' \
              % table_name
        else:
          if insert:
            hive_output.load.append(build_insert(insert, db_name, db_suffix, file_format,
                                        codec, compression_type, table_name, data_path,
                                        create_hive=create_hive))
          else:
              print 'Empty insert for table %s. Skipping insert generation' % table_name

    impala_output.write_to_file("load-%s-impala-generated-%s-%s-%s.sql" %
        (output_name, file_format, codec, compression_type))
    impala_load.write_to_file("load-%s-impala-load-generated-%s-%s-%s.sql" %
        (output_name, file_format, codec, compression_type))


  hive_output.write_to_file('load-' + output_name + '-hive-generated.sql')
  hbase_output.create.append("exit")
  hbase_output.write_to_file('load-' + output_name + '-hbase-generated.create')
  hbase_post_load.load.append("exit")
  hbase_post_load.write_to_file('post-load-' + output_name + '-hbase-generated.sql')

def parse_schema_template_file(file_name):
  VALID_SECTION_NAMES = ['DATASET', 'BASE_TABLE_NAME', 'COLUMNS', 'PARTITION_COLUMNS',
                         'ROW_FORMAT', 'CREATE', 'CREATE_HIVE', 'CREATE_KUDU',
                         'DEPENDENT_LOAD', 'LOAD',
                         'LOAD_LOCAL', 'ALTER', 'HBASE_COLUMN_FAMILIES', 'TABLE_PROPERTIES']
  return parse_test_file(file_name, VALID_SECTION_NAMES, skip_unknown_sections=False)

if __name__ == "__main__":
  if options.table_formats is None:
    if options.exploration_strategy not in KNOWN_EXPLORATION_STRATEGIES:
      print 'Invalid exploration strategy:', options.exploration_strategy
      print 'Valid values:', ', '.join(KNOWN_EXPLORATION_STRATEGIES)
      sys.exit(1)

    test_vectors = [vector.value for vector in\
        load_table_info_dimension(options.workload, options.exploration_strategy)]
  else:
    table_formats = options.table_formats.split(',')
    dataset = get_dataset_from_workload(options.workload)
    test_vectors =\
        [TableFormatInfo.create_from_string(dataset, tf) for tf in table_formats]

  target_dataset = test_vectors[0].dataset
  print 'Target Dataset: ' + target_dataset
  dataset_load_dir = os.path.join(SQL_OUTPUT_DIR, target_dataset)
  # If the directory containing the sql files does not exist, create it. Else nuke all the
  # files corresponding to the current workload.
  try:
    os.makedirs(dataset_load_dir)
  except OSError:
    # Directory already exists, remove it.
    shutil.rmtree(dataset_load_dir)
    # Recreate the workload dir
    os.makedirs(dataset_load_dir)
  finally:
    # Make sure that the directory was created and is empty.
    assert os.path.isdir(dataset_load_dir)
    assert len(os.listdir(dataset_load_dir)) == 0
    # Make the dataset dir the current working directory
    os.chdir(dataset_load_dir)

  schema_template_file = os.path.join(DATASET_DIR, target_dataset,
                                      '%s_schema_template.sql' % target_dataset)

  if not os.path.isfile(schema_template_file):
    print 'Schema file not found: ' + schema_template_file
    sys.exit(1)

  constraints_file = os.path.join(DATASET_DIR, target_dataset, 'schema_constraints.csv')
  include_constraints, exclude_constraints, only_constraints = \
      parse_table_constraints(constraints_file)
  sections = parse_schema_template_file(schema_template_file)
  generate_statements('%s-%s' % (options.workload, options.exploration_strategy),
      test_vectors, sections, include_constraints, exclude_constraints, only_constraints)