This repository has been archived by the owner. It is now read-only.

crystal-db-concurrency-showcase

Reproducing an issue with multiple concurrent queries to a database

crystal-db-concurrency-showcase

Reproducing an issue with multiple concurrent queries to a database as described in:

Prerequisites

  • Crystal (known to work with v.1.7)
  • Postgres server (known to work with v.14.7)
  • Optional: MySQL server (known to work with MariaDB v.10.6)
  • GNU make

Setup

TL;DR

make deps
make dotenv

# Now:
#  1. Tune your `.env` and `.env.test` files;
#  2. Create a database on Postgres (optionally, on MySQL).

make spec # optional sanity check
make
make db-seed

1. Install the dependencies

make deps

2. Create and tune .env* files

Create a .env (will also create a .env.test):

make dotenv

Tune the SHOWCASE_DB_URI variable in .env and .env.test to suite your environment.

Supported backends:

  • Postgres -- the default;
  • MySQL -- optional;
  • SQLite.

3. Create databases

For running on Postgres or MySQL, you need to create the development database (and optionally, the test database for running the specs).

Postgres:

On Debian-based Linux OS this should do it:

sudo -iu postgres
psql
# The Postgres prompt (postgres=#) should be seen
create database showcase_development;
create database showcase_test;
exit
exit

MySQL:

On Debian-based Linux OS this should do it:

mysql -u <dbuser> -p
> create database showcase_development;
> create database showcase_test;
> exit

<dbuser> is a MySQL user having the CREATE privilege (e.g. root).

SQLite:

No need to do anything, the database files would appear in ./db/ automatically. To drop a database, just delete the corresponding file.

4. Run the specs

This step is optional.

Doublecheck your .env.test settings. If you want to run the specs against more than one backend, you need to execute this step more than once, each time tuning you .env.test to point to a different database backend.

Run the specs:

make spec

5. Build the binaries

make

There are two binaries needed; this is going to build both:

  • bin/seed-tool -- used to seed the database;
  • bin/showcase -- the showcase app sending multiple concurrent requests to the database.

They both have a help screen shown with -h.

6. Seed the database

Doublecheck your .env settings. If you want to run the specs against more than one backend, you need to execute this step more than once, each time tuning you .env.test to point to a different database backend.

On decent hardware seeding takes:

  • ~3 min for Postgres;
  • ~2 min for MySQL/MariaDB;
  • ~15 min for SQLite;

To do the seeding:

make db-seed

Run the showcase app

You should have completed the Setup procedure explained above to run the showcase app. Doublecheck your .env file.

Running the showcase app against a Postgres db backend prepared as explained above should reveal the issue with the linearly growing query round trip times. A hundred concurrent queries should demonstrate it:

bin/showcase -c100 # 100 almost simultaneous queries

The available options are explained in the help page:

bin/showcase -h

Review the output

The issue we want to resolve is described here.

For Postgres, a typical 100 queries run would look like

I: Spawning 100 requests...
I: Received 1 (idx=46) team_id=1 size=44000 time=0.3s
I: Received 2 (idx=84) team_id=2 size=33000 time=0.32s
I: Received 3 (idx=87) team_id=3 size=33000 time=0.34s
.
.
I: Received 98 (idx=3) team_id=2 size=33000 time=6.47s
I: Received 99 (idx=39) team_id=3 size=33000 time=6.54s
I: Received 100 (idx=23) team_id=1 size=44000 time=6.63s
I: Total time: 7.7s

Sometimes fibers seem to get lost and the routine collecting the results hangs while waiting for all of them to come back:

.
.
I: Received 96 (idx=38) team_id=1 size=44000 time=5.93s
I: Received 97 (idx=85) team_id=1 size=44000 time=6.01s
I: Received 98 (idx=11) team_id=3 size=33000 time=6.84s
[hanging here...]

and you need to interrupt this with <Ctrl-C>.

The same run against the SQLite backend would look different:

.
.
I: Received 98 (idx=99) team_id=3 size=33000 time=0.03s
I: Received 99 (idx=70) team_id=3 size=33000 time=0.03s
I: Received 100 (idx=59) team_id=2 size=33000 time=0.03s
I: Total time: 3.0s

The times per query processing do not grow and all fibers eventually come back.

Experiment show that on MySQL the times grow similarly to the Postgres backend and often some fibers get lost as well.

Development

Changes helping solve the issue are more than welcome, see "Contributing".

Contributing

  1. Fork it (https://github.com/dext/crystal-db-concurrency-showcase/fork);
  2. Create your feature branch -- git checkout -b my-change
  3. Commit your changes -- git commit -am 'Add the change'
  4. Push to the branch -- git push origin my-change
  5. Create a new Pull Request.

Contact us

  • yassen.damyanov -AT- dext.com
Repository

crystal-db-concurrency-showcase

Owner
Statistic
  • 0
  • 0
  • 0
  • 0
  • 4
  • over 1 year ago
  • March 21, 2023
License

MIT License

Links
Synced at

Sun, 17 Nov 2024 12:17:49 GMT

Languages