Add tooling to gather TPC-C transactional data

tools/calibration/transactions/chbenchmark/COND

@@ -0,0 +1,44 @@
+RUN_FOR_S = 3 * 60 + 30  # 3 minutes 30 seconds
+NUM_CLIENTS = {
+  "db.r6g.xlarge": [1, 2, 4, 8, 16, 32, 64],
+  "db.r6g.2xlarge": [1, 2, 4, 8, 16, 32, 64, 128],
+  "db.r6g.4xlarge": [1, 2, 4, 8, 16, 32, 64, 128, 256],
+  "db.r6g.large": [1, 2, 4, 8, 16, 32],
+  "db.t4g.medium": [1, 2, 4, 8, 16, 32],
+}
+
+INSTANCES = list(NUM_CLIENTS.keys())
+
+COND_INSTANCES = {
+  instance: instance.replace(".", "_").replace("db.", "") for instance in INSTANCES
+}
+
+combine(
+  name="all",
+  deps=[
+    ":{}".format(COND_INSTANCES[instance])
+    for instance in INSTANCES
+  ],
+)
+
+for instance in INSTANCES:
+  run_experiment_group(
+    name=COND_INSTANCES[instance],
+    run="./run_instance.sh",
+    experiments=[
+      ExperimentInstance(
+        name="{}-{}".format(COND_INSTANCES[instance], clients),
+        options={
+          "t-clients": clients,
+          "run-for-s": RUN_FOR_S,
+          "system-config-file": "system_config_chbench.yml",
+          "physical-config-file": "../../../../config/physical_config_chbench.yml",
+          "txn-warehouses": 1740,
+          "txn-config-file": "aurora.config",
+          "schema-name": "chbenchmark",
+          "instance": instance,
+        },
+      )
+      for clients in NUM_CLIENTS[instance]
+    ],
+  )

tools/calibration/transactions/chbenchmark/retrieve_metrics.py

@@ -0,0 +1,117 @@
+import argparse
+import asyncio
+from datetime import timedelta
+from typing import List
+
+from brad.config.file import ConfigFile
+from brad.provisioning.directory import Directory
+from brad.daemon.metrics_def import MetricDef
+from brad.daemon.perf_insights import PerfInsightsClient
+
+
+BASE_METRICS = [
+    "os.loadAverageMinute.one",
+    "os.loadAverageMinute.five",
+    "os.loadAverageMinute.fifteen",
+    "os.cpuUtilization.system",
+    "os.cpuUtilization.total",
+    "os.cpuUtilization.user",
+    "os.diskIO.avgQueueLen",
+    "os.diskIO.tps",
+    "os.diskIO.util",
+    "os.diskIO.readIOsPS",
+    "os.diskIO.readKbPS",
+    "os.diskIO.writeIOsPS",
+    "os.diskIO.writeKbPS",
+    "os.network.rx",
+    "os.network.tx",
+    "os.memory.active",
+    "os.memory.dirty",
+    "os.memory.free",
+    "os.memory.writeback",
+    "os.memory.total",
+    "os.tasks.blocked",
+    "os.tasks.running",
+    "os.tasks.sleeping",
+    "os.tasks.stopped",
+    "os.tasks.total",
+    "db.SQL.queries",
+    "db.SQL.total_query_time",
+    "db.SQL.tup_deleted",
+    "db.SQL.tup_fetched",
+    "db.SQL.tup_inserted",
+    "db.SQL.tup_returned",
+    "db.SQL.tup_updated",
+    "db.Transactions.active_transactions",
+    "db.Transactions.blocked_transactions",
+    "db.Transactions.duration_commits",
+    "db.Transactions.xact_commit",
+    "db.Transactions.xact_rollback",
+    # NOTE: Aurora has specific storage metrics (probably because they use a custom storage engine)
+    # https://docs.aws.amazon.com/AmazonRDS/latest/AuroraUserGuide/USER_PerfInsights_Counters.html#USER_PerfInsights_Counters.Aurora_PostgreSQL
+    "os.diskIO.auroraStorage.auroraStorageBytesRx",
+    "os.diskIO.auroraStorage.auroraStorageBytesTx",
+    "os.diskIO.auroraStorage.diskQueueDepth",
+    "os.diskIO.auroraStorage.readThroughput",
+    "os.diskIO.auroraStorage.writeThroughput",
+    "os.diskIO.auroraStorage.readLatency",
+    "os.diskIO.auroraStorage.writeLatency",
+    "os.diskIO.auroraStorage.readIOsPS",
+    "os.diskIO.auroraStorage.writeIOsPS",
+]
+
+ALL_METRICS: List[MetricDef] = []
+for m in BASE_METRICS:
+    # N.B. The metrics are reported no more than once a minute. So
+    # average/max/min will all report the same number.
+    ALL_METRICS.append((m, "avg"))
+
+
+def main():
+    parser = argparse.ArgumentParser()
+    parser.add_argument(
+        "--instance-id",
+        type=str,
+        help="The Aurora instance's identifier.",
+    )
+    parser.add_argument(
+        "--physical-config-file",
+        type=str,
+        help="Used to specify the Aurora instance instead of by an ID.",
+    )
+    parser.add_argument(
+        "--out-file",
+        type=str,
+        required=True,
+        help="The path where the results should be saved.",
+    )
+    parser.add_argument(
+        "--num-prev-points",
+        type=int,
+        default=60,
+        help="The number of metric data points to retrieve.",
+    )
+    args = parser.parse_args()
+
+    if args.instance_id is not None:
+        client = PerfInsightsClient.from_instance_identifier(
+            instance_identifier=args.instance_id
+        )
+    elif args.physical_config_file is not None:
+        config = ConfigFile.load_from_physical_config(args.physical_config_file)
+        directory = Directory(config)
+        asyncio.run(directory.refresh())
+        client = PerfInsightsClient(resource_id=directory.aurora_writer().resource_id())
+    else:
+        raise RuntimeError()
+
+    metrics = client.fetch_metrics(
+        ALL_METRICS,
+        period=timedelta(minutes=1),
+        num_prev_points=args.num_prev_points,
+    )
+    metrics.to_csv(args.out_file)
+
+
+if __name__ == "__main__":
+    main()

tools/calibration/transactions/chbenchmark/run_instance.sh

@@ -0,0 +1,84 @@
+#! /bin/bash
+
+function extract_named_arguments() {
+  # Evaluates any environment variables in this script's arguments. This script
+  # should only be run on trusted input.
+  orig_args=($@)
+  for val in "${orig_args[@]}"; do
+    phys_arg=$(eval "echo $val")
+
+    if [[ $phys_arg =~ --t-clients=.+ ]]; then
+      t_clients=${phys_arg:12}
+    fi
+
+    if [[ $phys_arg =~ --run-for-s=.+ ]]; then
+      run_for_s=${phys_arg:12}
+    fi
+
+    if [[ $phys_arg =~ --system-config-file=.+ ]]; then
+      system_config_file=${phys_arg:21}
+    fi
+
+    if [[ $phys_arg =~ --physical-config-file=.+ ]]; then
+      physical_config_file=${phys_arg:23}
+    fi
+
+    if [[ $phys_arg =~ --txn-warehouses=.+ ]]; then
+      txn_warehouses=${phys_arg:17}
+    fi
+
+    if [[ $phys_arg =~ --txn-config-file=.+ ]]; then
+      txn_config_file=${phys_arg:18}
+    fi
+
+    if [[ $phys_arg =~ --schema-name=.+ ]]; then
+      schema_name=${phys_arg:14}
+    fi
+
+    if [[ $phys_arg =~ --instance=.+ ]]; then
+      instance=${phys_arg:11}
+    fi
+  done
+}
+
+script_loc=$(cd $(dirname $0) && pwd -P)
+cd $script_loc
+extract_named_arguments $@
+
+abs_txn_config_file=$(realpath $txn_config_file)
+abs_physical_config_file=$(realpath $physical_config_file)
+abs_system_config_file=$(realpath $system_config_file)
+
+>&2 echo "Adjusting blueprint"
+brad admin --debug modify_blueprint \
+  --schema-name $schema_name \
+  --physical-config-file $abs_physical_config_file \
+  --system-config-file $abs_system_config_file \
+  --aurora-instance-type $instance \
+  --aurora-num-nodes 1
+
+>&2 echo "Waiting 30 seconds before retrieving pre-metrics..."
+sleep 30
+
+>&2 echo "Retrieving pre-metrics..."
+python3 retrieve_metrics.py --out-file $COND_OUT/pi_metrics_before.csv --physical-config-file $abs_physical_config_file
+
+>&2 echo "Running the transactional workload..."
+
+# We run against Aurora directly.
+pushd ../../../../workloads/chbenchmark/py-tpcc/
+RECORD_DETAILED_STATS=1 python3 -m pytpcc.tpcc aurora \
+  --no-load \
+  --config $abs_txn_config_file \
+  --warehouses $txn_warehouses \
+  --duration $run_for_s \
+  --clients $t_clients \
+  --scalefactor 1 \
+  --lat-sample-prob 0.25
+popd
+
+>&2 echo "Waiting 10 seconds before retrieving metrics..."
+sleep 10
+
+>&2 echo "Retrieving metrics..."
+python3 retrieve_metrics.py --out-file $COND_OUT/pi_metrics.csv --physical-config-file $abs_physical_config_file