--db-pgbench and --workload-pgbench

[michelp]

No description provided.

[NikolayS]

Awesome @michelp!

Checked, worked like a charm.

We could also catch pgbench's output (where it prints tps and so on) into a file, considering it as a new artifact and delivering to the artifacts directory, but this is definitely could be a separate PR.

This – having pgbench on board – can help with benchmarking AWS instances and checking various values for Postgres parameters, now it's so simple!

[NikolayS]

Cool:

nancy run \
  --db-pgbench "-s 100" \
  --workload-pgbench "-t 60 -j 16 -c 16" \
  --run-on aws  \
  --aws-ec2-type "i3.large"  \
  --aws-keypair-name awskey \
  --aws-ssh-key-path file://$(echo ~)/.ssh/awskey.pem

results:

transaction type: <builtin: TPC-B (sort of)>
scaling factor: 100
query mode: simple
number of clients: 16
number of threads: 16
number of transactions per client: 60
number of transactions actually processed: 960/960
latency average = 20.040 ms
tps = 798.414235 (including connections establishing)
tps = 817.377725 (excluding connections establishing)
...
------------------------------------------------------------------------------
Artifacts (collected in "./nancy-20180920-003245N-MSK/"):
  Postgres config:    postgresql.conf
  Postgres logs:      postgresql.prepare.log.gz (preparation),
                      postgresql.workload.log.gz (workload)
  pgBadger reports:   pgbadger.html (for humans),
                      pgbadger.json (for robots)
  Stat stapshots:     pg_stat_statements.csv,
                      pg_stat_***.csv
------------------------------------------------------------------------------
Total execution time: 0:05:28
------------------------------------------------------------------------------
Workload:
  Execution time:     0:00:03
  Total query time:   848.31000000001  ms
  Queries:            4801
  Query groups:       6
  Errors:             0
  Errors groups:      0
------------------------------------------------------------------------------

The answering to questions like "will an instance with X CPU / Y GB RAM and Postgres 10 handle *** INSERTs per seconds?" is SO easy now.