SQL Jobber

sql-jobber is a light weight SQL "job server" that maintains a distributed, asynchronous job queue of SQL read jobs against one or more large source databases. The results are written to one or more separate "cache" databases, where each result set is a newly created table, from where the results can be fetched as many times much faster than querying the source databases.

This is useful for queuing and offloading report generation on applications without clogging source databases, especially in the case of end user applications.

Parts

• Supports MySQL and PostgreSQL as source databases.
• Supports MySQL and PostgreSQL as result / cache databases for job responses.
• Standalone server that exposes HTTP APIs for managing jobs and groups of jobs (list, post, status check, cancel jobs).
• Reads SQL queries from .sql files and registers them as jobs ready to be queued.
• Written in Go and built on top of Machinery. Supports multi-process, multi-threaded, asynchronous distributed job queueing via a common broker backend (Redis, AMQP etc.)

Usecase

An application that has a very large SQL database, when there are several thousand concurrent users requesting reports from the database simultaneously, every second of IO delay in query execution locks up the application's threads, snowballing and overloading the application.

Instead, defer every single report request into a job queue, there by immediately freeing up the front end application. The reports are presented to users as they're executed (frontend polls the job's status and prevents the user from sending any more queries). Fixed SQL Jobber servers and worker threads also act as traffic control and prevent the primary database from being indundated with requests.

Once the reports are generated, it's natural for users to further transform the results by slicing, sorting, and filtering, generating additional queries to the primary database. To offset this load, these subsequent queries can be sent to the smaller, much faster results cache database. These results are of course ephemeral and can be thrown away or expired.

Concepts

Task

A task is a named SQL job is loaded into the server on startup. Tasks are defined in .sql files in the simple goyesql format. Such queries are self-contained and produce the desired final output with neatly named columns. They can take arbitrary positional arguments for execution. A task can be attached to one or more specific databases defined in the configuration using the -- db: tag. In case of multiple databases, the query will be executed against a random one from the list, unless a specific database is specified in the API request (db). A -- queue: tag to always route the task to a particular queue, unless it's overriden by the queue param when making a job request. A -- results: tag specifies the results backend to which the results of a task will be written. If there are multiple result backends specified, the results are written a random one.

Example:

-- queries.sql

-- name: get_profit_summary
SELECT SUM(amount) AS total, entry_date FROM entries GROUP BY entry_date WHERE user_id = ?;

-- name: get_profit_entries
-- db: db0, other_db
-- queue: myqueue
-- results: my_res_db
SELECT * FROM entries WHERE user_id = ?;

-- name: get_profit_entries_by_date
SELECT * FROM entries WHERE user_id = ? AND timestamp > ? and timestamp < ?;

-- name: get_profit_entries_by_date
-- raw: 1
-- This query will not be prepared (raw=1)
SELECT * FROM entries WHERE user_id = ? AND timestamp > ? and timestamp < ?;

Here, when the server starts, the queries get_profit_summary and get_profit_entries are registered automatically as tasks. Internally, the server validates and prepares these SQL statements (unless raw: 1). ? are MySQL value placholders. For Postgres, the placeholders are $1, $2 ...

Job

A job is an instance of a named task that has been queued to run. Each job has an ID that can be used to track its status. If an ID is not passed explicitly, it is generated internally and returned. These IDs need not be unique, but only a single job with a certain ID can run at any given point. For the next job with the same ID to be scheduled, the previous job has to finish execution. Using non-unique IDs like this is useful in cases where users can be prevented from sending multiple requests for the same reports, like in our usecases.

An application polls with the job ID to check if results are ready for consumption.

Results

The results from an SQL query job are written to a result backend (MySQL or Postgres) from where they can be further read or queried. This is configured in the configuration file. The results from a job are written to a new table named after that job, where schema of the results table is automatically generated from the results of the original SQL query. All fields are transformed into one of these types BIGINT, DECIMAL, TIMESTAMP, DATE, BOOLEAN, TEXT.

Installation

A pre-compiled binary can be downloaded from the releases page.

2) Configure

Copy the config.toml.sample file as config.toml somewhere and edit the configuration values.

3) Setup tasks

Write your SQL query tasks in .sql files in the goyesql format (as shown in the examples earlier) and put them in a directory somewhere.

4) Start the server

sql-jobber --config /path/to/config.toml --sql-directory /path/to/your/sql/queries

# Run 'sql-jobber --help' to see all supported arguments

Starting the server runs a set of workers listening on a default job queue. It also starts an HTTP service on http://127.0.0.1:6060 which is the control interface. It's possible to run the server without the HTTP interface by passing the --worker-only flag.

Usage

POST requests accept raw JSON bodies. The JSON params are listed below.

Schedule a job

$ curl localhost:6060/tasks/get_profit_entries_by_date/jobs -H "Content-Type: application/json" -X POST --data '{"job_id": "myjob", "args": ["USER1", "2017-12-01", "2017-01-01"]}'

{"status":"success","data":{"job_id":"myjob","task_name":"get_profit_entries_by_date","queue":"sqljob_queue","eta":null,"retries":0}}

Schedule a group of jobs

Sometimes, it's necessary to schedule a group of jobs and perform an action once they're all complete. Group jobs here run concurrently and independent of each other. The group state can be polled to figure out if all the jobs in it have finished executing.

$ curl localhost:6060/groups -H "Content-Type: application/json" -X POST --data '{"group_id": "mygroup", "concurrency": 3, "jobs": [{"job_id": "myjob", "task": "get_profit_entries_by_date", "args": ["USER1", "2017-12-01", "2017-01-01"]}, {"job_id": "myjob2", "task": "get_profit_entries_by_date", "args": ["USER1", "2017-12-01", "2017-01-01"]}]'

{"status":"success","data":{"group_id":"mygroup","jobs":[{"job_id":"myjob","task":"test1","queue":"sqljob_queue","eta":null,"retries":0},{"job_id":"myjob2","task":"test2","queue":"sqljob_queue","eta":null,"retries":0}]}}

Check a job's status

$ curl localhost:6060/jobs/myjob
{"status":"success","data":{"job_id":"myjob","status":"SUCCESS","results":[{"Type":"int64","Value":2}],"error":""}}~                                                                               

# `Results` indicates the number of rows generated by the query.

Advanced usage

Multiple queues, workers, and job distribution

It's possible to run multiple workers on one or more machines that run different jobs with different concurrency levels independently of each other using different queues. Not all of these instances need to expose the HTTP service and can run as --worker-only. This doesn't really make a difference as long as all instances connect to the same broker backend. A job posted to any instance will be routed correctly to the right instances based on the queue parameter.

Often times, different queries have different priorities of execution. Some may need to return results faster than others. The below example shows two SQL Jobber servers being run, one with 30 workers and one with just 5 to process jobs of different priorities.

# Run the primary worker + HTTP control interface
sql-jobber --config /path/to/config.toml --sql-directory /path/to/sql/dir \
	--queue "high_priority" \
    --worker-name "high_priority_worker" \
    --worker-concurrency 30

# Run another worker on a different queue to handle low priority jobs
sql-jobber --config /path/to/config.toml --sql-directory /path/to/sql/dir \
	--queue "low_priority" \
    --worker-name "low_priority_worker" \
    --worker-concurrency 5 \
    --worker-only

# Send a job to the high priority queue.
$ curl localhost:6060/tasks/get_profit_entries_by_date/jobs -H "Content-Type: application/json" --data '{"job_id": "myjob", "queue": "high_priority", "args": ["USER1", "2017-12-01", "2017-01-01"]}'

# Send another job to the low priority queue.
$ curl localhost:6060/tasks/get_profit_entries_by_date/jobs -H "Content-Type: application/json" --data '{"job_id": "myjob", "queue": "low_priority"}'

API client

github.com/knadh/sql-jobber/client package can be used as a Go HTTP API client for sql-jobber.

License

Licensed under the MIT License.