BigQuery Benchmark
This directory contains benchmark scripts for BigQuery client. It is created primarily for project maintainers to measure library performance.
Usage
python benchmark.py
Flags
Run python benchmark.py -h for detailed information on available flags.
--reruns can be used to override the default number of times a query is rerun. Must be a positive
integer. Default value is 3.
--projectid can be used to run benchmarks in a different project. If unset, the GOOGLE_CLOUD_PROJECT
environment variable is used.
--queryfile can be used to override the default file which contains queries to be instrumented.
--table can be used to specify a table to which benchmarking results should be streamed. The format
for this string is in BigQuery standard SQL notation without escapes, e.g. projectid.datasetid.tableid
--create_table can be used to have the benchmarking tool create the destination table prior to streaming.
--tag allows arbitrary key:value pairs to be set. This flag can be specified multiple times.
When --create_table flag is set, must also specify the name of the new table using --table.
Example invocations
Setting all the flags
python benchmark.py \
--reruns 5 \
--projectid test_project_id \
--table logging_project_id.querybenchmarks.measurements \
--create_table \
--tag source:myhostname \
--tag somekeywithnovalue \
--tag experiment:special_environment_thing
Or, a more realistic invocation using shell substitions:
python benchmark.py \
--reruns 5 \
--table $BENCHMARK_TABLE \
--tag origin:$(hostname) \
--tag branch:$(git branch --show-current) \
--tag latestcommit:$(git log --pretty=format:'%H' -n 1)
Stream Results To A BigQuery Table
When streaming benchmarking results to a BigQuery table, the table schema is as follows:
[
{
"name": "groupname",
"type": "STRING"
},
{
"name": "name",
"type": "STRING"
},
{
"name": "tags",
"type": "RECORD",
"mode": "REPEATED",
"fields": [
{
"name": "key",
"type": "STRING"
},
{
"name": "value",
"type": "STRING"
}
]
},
{
"name": "SQL",
"type": "STRING"
},
{
"name": "runs",
"type": "RECORD",
"mode": "REPEATED",
"fields": [
{
"name": "errorstring",
"type": "STRING"
},
{
"name": "start_time",
"type": "TIMESTAMP"
},
{
"name": "query_end_time",
"type": "TIMESTAMP"
},
{
"name": "first_row_returned_time",
"type": "TIMESTAMP"
},
{
"name": "all_rows_returned_time",
"type": "TIMESTAMP"
},
{
"name": "total_rows",
"type": "INTEGER"
}
]
},
{
"name": "event_time",
"type": "TIMESTAMP"
}
]
The table schema is the same as the benchmark in go, so results from both languages can be streamed to the same table.