[Benchmark] Refactor benchmark system.

Instead of using a different script for each experiment and DBMS to
benchmark, implement `connectors` to these DBMS's. The connector has a
method to execute an experiment with the given parameters and returns
the measured times.

In addition, the format of the YAML files of the experiments has been
refactored to contain all the information and parameters to execute them
on each connector.

`Benchmark.py` is refactored as well to read the experiment files and
execute them on each available specified connector, with possibly
multiple configurations.

Some more minor changes:
- Benchmark script now has the option to execute one (or multiple)
  specific experiments.
- The `run_id` of each experiment run is tracked and inserted into the
  database.

benchmark/_schema.yml

@@ -1,28 +1,19 @@
 description: str()
-version: int(required=False, min=1)
 suite: str()
 benchmark: str()
 name: str(required=False)
 readonly: bool()
-pattern: str()
-args: str(required=False)
-configurations: map(str(), required=False)
-tables: include('table_list')
-cases: map(any(str(), include('case')), key=any())
-compare_to: map(required=False)
-chart: include('chart', required=False)
+chart: include('chart_def', required=False)
+data: map(include('table'), required=False)      # Map from table name to 'table'
+systems:
+    mutable: include('mutable', required=False)
+    PostgreSQL: include('PostgreSQL', required=False)
+    DuckDB: include('DuckDB', required=False)
+    HyPer: include('HyPer', required=False)
 ---
-table:
-    name: str()                             # table name
-    path: str(required=False)               # path to table file
-    sf: num(required=False, min=0, max=1)   # scale factor as portion of file to load; defaults to 1
-    delimiter: str(required=False)          # defaults to ','
-    header: int(required=False)             # 1 if file has header, 0 otherwise; defaults to 0
-table_list:
-    list(any(str(), include('table')), required=True)
-case:
-    query: str()
-    tables: include('table_list')
+chart_def:
+    x: include('axis', required=False)
+    y: include('axis', required=False)
 ---
 axis:
     # Kind of scale, one of
@@ -37,6 +28,28 @@ axis:
     type: str(required=False)
     # A label for the axis
     label: str(required=False)
-chart:
-    x: include('axis', required=False)
-    y: include('axis', required=False)
+---
+table:
+    attributes: map(str(), key=str(), required=False)           # table column names and types
+    file: str(required=False)                   # path to table file
+    delimiter: str(required=False)              # defaults to ','
+    header: int()                               # 1 if file has header, 0 otherwise; defaults to 0
+    format: str(required=False)                 # file format
+    scale_factors: map(num(min=0, max=1), required=False) # map from case name to scale factor (portion of file to load)
+    lines_in_file: int(required=False)          # Number of lines in the file. Is counted and added by the benchmark script
+---
+mutable:
+    cases: include('cases')
+    pattern: str()
+    args: str(required=False)
+    configurations: map(str(), required=False)
+PostgreSQL:
+    cases: include('cases')
+DuckDB:
+    cases: include('cases')
+HyPer:
+    single_core: bool(required=False)
+    all_cores: bool(required=False)
+    cases: include('cases')
+---
+cases: map(str(), key=any())

benchmark/database_connectors/connector.py

@@ -0,0 +1,27 @@
+from abc import ABC, abstractmethod
+
+class Connector(ABC):
+
+    # Function that performs an experiment n_runs times given the parameters `params`.
+    # Returns a dict with the measured times for the experiment and configuration.
+    # Result has the form:
+    # results
+    # └── configurations
+    #     └── cases
+    #         └── times (list)
+    #
+    # results:          configuration name      --> configuration
+    # configuration:    case  --> times
+    # times:            list of floats (size=n_runs)
+    #
+    # Example: (n_runs=2)
+    #   {
+    #       'PostgreSQL':
+    #           1: [1235.093, 1143.43],
+    #           2: [1033.711, 1337.37],
+    #           3: [1043.452, 1010.01],
+    #           4: [1108.702, 1234.56]
+    #   }
+    @abstractmethod
+    def execute(self, n_runs: int, params: dict):
+        pass

benchmark/database_connectors/duckdb.py

@@ -0,0 +1,170 @@
+from .connector import *
+
+import duckdb
+import os
+import json
+
+
+TMP_DB = 'tmp.duckdb'
+TMP_SQL_FILE = 'tmp.sql'
+
+# TODO way of measuring time is wrong. Use duckdb_cli like in older version.
+
+class DuckDB(Connector):
+
+    def __new__(cls, *args, **kwargs):
+        return super().__new__(cls)
+
+
+    def __init__(self, duckdb_cli, verbose=False):
+        self.duckdb_cli=duckdb_cli
+
+
+    # Runs an experiment 'n_runs' times, all parameters are in 'params'
+    def execute(self, n_runs, params: dict):
+        self.clean_up()
+
+        measurement_times = dict()      # map that is returned with the measured times
+
+        # Check wether tables contain scale factors
+        with_scale_factors = False
+        for table in params['data'].values():
+            if (table.get('scale_factors')):
+                with_scale_factors = True
+                break
+
+        for _ in range(n_runs):
+            try:
+                # Set up database
+                self.generate_create_table_stmts(params['data'], with_scale_factors)
+
+
+                # If tables contain scale factors, they have to be loaded separately for every case
+                if (with_scale_factors and bool(params.get('readonly'))):
+                    # Write cases/queries to a file that will be passed to the command to execute
+                    statements = list()
+                    for case, query_stmt in params['cases'].items():
+                        # Create tables from tmp tables with scale factor
+                        for table_name, table in params['data'].items():
+                            statements.append(f"DELETE FROM {table_name};")     # empty existing table
+                            if table.get('scale_factors'):
+                                sf = table['scale_factors'][case]
+                            else:
+                                sf = 1
+                            header = int(table.get('header', 0))
+                            num_rows = round((table['lines_in_file'] - header) * sf)
+                            statements.append(f"INSERT INTO {table_name} SELECT * FROM {table_name}_tmp LIMIT {num_rows};")
+
+                        statements.append(".timer on")
+                        statements.append(query_stmt)   # Actual query from this case
+                        statements.append(".timer off")
+
+                    # Append statements to file
+                    with open(TMP_SQL_FILE, "a+") as tmp:
+                        for stmt in statements:
+                            tmp.write(stmt + "\n")
+
+
+
+                # Otherwise, tables have to be created just once before the measurements (done above)
+                else:
+                    # Write cases/queries to a file that will be passed to the command to execute
+                    with open(TMP_SQL_FILE, "a+") as tmp:
+                        tmp.write(".timer on\n")
+                        for case_query in params['cases'].values():
+                            tmp.write(case_query + '\n')
+                        tmp.write(".timer off\n")
+
+
+                # Execute query file and collect measurement data
+                command = f"./{self.duckdb_cli} {TMP_DB} < {TMP_SQL_FILE}" + " | grep 'Run Time' | cut -d ' ' -f 5 | awk '{print $1 * 1000;}'"
+                stream = os.popen(f'{command}')
+                for idx, line in enumerate(stream):
+                    time = float(line.replace("\n", "").replace(",", ".")) # in milliseconds
+                    case = list(params['cases'].keys())[idx]
+                    if case not in measurement_times.keys():
+                        measurement_times[case] = list()
+                    measurement_times[case].append(time)
+                stream.close()
+
+
+            finally:
+                self.clean_up()
+
+        return {'DuckDB': measurement_times}
+
+
+    # Deletes the used temporary database
+    def clean_up(self):
+        if os.path.exists(TMP_DB):
+            os.remove(TMP_DB)
+        if os.path.exists(TMP_SQL_FILE):
+            os.remove(TMP_SQL_FILE)
+
+
+    # Parse attributes of one table, return as string
+    def parse_attributes(self, attributes: dict):
+        columns = '('
+        for column_name, ty in attributes.items():
+            not_null = 'NOT NULL' if 'NOT NULL' in ty else ''
+            ty = ty.split(' ')
+            match (ty[0]):
+                case 'INT':
+                    typ = 'INT'
+                case 'CHAR':
+                    typ = f'CHAR({ty[1]})'
+                case 'DECIMAL':
+                    typ = f'DECIMAL({ty[1]},{ty[2]})'
+                case 'DATE':
+                    typ = 'DATE'
+                case 'DOUBLE':
+                    typ = 'DOUBLE'
+                case 'FLOAT':
+                    typ = 'REAL'
+                case 'BIGINT':
+                    typ = 'BIGINT'
+                case _:
+                    raise Exception(f"Unknown type given for '{column_name}'")
+            columns += f"{column_name} {typ} {not_null}, "
+        columns = columns[:-2] + ')'
+        return columns
+
+
+    # Creates tables in the database and copies contents of given files into them
+    # Call with 'with_scale_factors'=False if data should be loaded as a whole
+    # Call with 'with_scale_factors'=True if data should be placed in tmp tables
+    # and copied for each case with different scale factor
+    def generate_create_table_stmts(self, data: dict, with_scale_factors):
+        statements = list()
+        for table_name, table in data.items():
+            columns = self.parse_attributes(table['attributes'])
+
+            delimiter = table.get('delimiter')
+            header = table.get('header')
+            format = table['format'].upper()
+
+            if with_scale_factors:
+                table_name += "_tmp"
+
+            create = f"CREATE TABLE {table_name} {columns};"
+            copy = f"COPY {table_name} FROM '{table['file']}' ( "
+            if delimiter:
+                delim = delimiter.replace("'", "")
+                copy += f" DELIMITER \'{delim}\',"
+            if format:
+                copy += f" FORMAT {format},"
+            if header:
+                copy += f" HEADER," if (header==1) else ""
+
+            copy = copy[:-1] + " );"
+
+            statements.append(create)
+            statements.append(copy)
+
+            if with_scale_factors:
+                # Create actual table that will be used for experiment
+                statements.append(f"CREATE TABLE {table_name[:-4]} {columns};")
+
+        with open(TMP_SQL_FILE, "w") as tmp:
+            for stmt in statements:
+                tmp.write(stmt + "\n")