bulker Documentation

1. Project Overview & Quickstart (jitsucom/bulker) Copy Chapter

File: README.md Copy Chapter

[!CAUTION]

🚨 This repository has been archived Copy Chapter

Bulker has moved. The code now lives inside the main Jitsu monorepo:
👉 jitsucom/jitsu — /bulker

This repo is read-only. Please open issues and PRs in the main repo.

🚚 Bulker

Bulker is a tool for streaming and batching large amount of semi-structured data into data warehouses. It uses Kafka internally

How it works? Copy Chapter

Send and JSON object to Bulker HTTP endpoint, and it will make sure it will be saved to data warehouse:

• JSON flattening. Your object will be flattened - {a: {b: 1}} becomes {a_b: 1}
• Schema managenent for semi-structured data. For each field, bulker will make sure that a corresponding column exist in destination table. If not, Bulker
  will create it. Type will be best-guessed by value, or it could be explicitely set via type hint as in {"a": "test", "__sql_type_a": "varchar(4)"}
• Reliability. Bulker will put the object to Kafka Queue immediately, so if datawarehouse is down, data won't be lost
• Streaming or Batching. Bulker will send data to datawarehouse either as soon it become available in Kafka (streaming) or after some time (batching). Most
  data warehouses won't tolerate large number of inserts, that's why we implemented batching

Bulker is a 💜 of Jitsu, an open-source data integration platform.

See full list of features below

Bulker is also available as a go library if you want to embed it into your application as opposed to use a HTTP-server

Features Copy Chapter

• 🛢️ Batching - Bulker sends data in batches in most efficient way for particular database. For example, for Postgres it uses
  COPY command, for BigQuery it uses batch-files
• 🚿 Streaming - alternatively, Bulker can stream data to database. It is useful when number of records is low. Up to 10 records
  per second for most databases
• 🐫 Deduplication - if configured, Bulker will deduplicate records by primary key
• 📋 Schema management - Bulker creates tables and columns on the fly. It also flattens nested JSON-objects. Example if you send {"a": {"b": 1}} to
  bulker, it will make sure that there is a column a_b in the table (and will create it)
• 🦾 Implicit typing - Bulker infers types of columns from JSON-data.
• 📌 Explicit typing - Explicit types can be by type hints that are placed in JSON. Example: for event {"a": "test", "__sql_type_a": "varchar(4)"}
  Bulker will make sure that there is a column a, and it's type is varchar(4).
• 📈 Horizontal Scaling. Bulker scales horizontally. Too much data? No problem, just add Bulker instances!
• 📦 Dockerized - Bulker is dockerized and can be deployed to any cloud provider and k8s.
• ☁️ Cloud Native - each Bulker instance is stateless and is configured by only few environment variables.

Supported databases Copy Chapter

Bulker supports the following databases:

• ✅ PostgresSQL
• ✅ Redshit
• ✅ Snowflake
• ✅ Clickhouse
• ✅ BigQuery
• ✅ MySQL
• ✅ S3
• ✅ GCS

Please see Compatibility Matrix to learn what Bulker features are supported by each database.

Documentation Links Copy Chapter

Note
We highly recommend to read Core Concepts below before diving into details

• How to use Bulker as HTTP Service
  • Server Configuration
  • HTTP API
• How to use bulker as Go-lib (coming soon)

Core Concepts Copy Chapter

Destinations Copy Section

Bulker operates with destinations. Destination is database or
storage service (e.g. S3, GCS). Each destination has an ID and configuration
which is represented by JSON object.

Bulker exposes HTTP API to load data into destinations, where those
destinations are referenced by their IDs.

If destination is a database, you'll need to provide a destination table name.

Event Copy Section

The main unit of data in Bulker is an event. Event is a represented JSON-object

Batching and Streaming (aka Destination Mode) Copy Section

Bulker can send data to database in two ways:

• Streaming. Bulker sends evens to destinaion one by one. It is useful when number of events is low (less than 10 events per second for most DBs).
• Batching. Bulker accumulates events in batches and sends them periodically once batch is full or timeout is reached. Batching is more efficient for large amounts of events. Especially for cloud data-warehouses
  (e.g. Postgres, Clickhouse, BigQuery).

Primary Keys and Deduplication Copy Section

Optionally, Bulker can deduplicate events by primary key. It is useful when you same event can be sent to Bulker multiple times.
If available, Bulker uses primary keys, but for some data warehouses alternative strategies are used.

Read more about deduplication »

File: connectors/airbytecdk/README.md Copy Chapter

Airbyte - Golang SDK/CDK

This package aims to help developers build connectors (sources/destinations) really fast in Go.
The focus of this package is developer efficiency. It focusses on letting developers focus more on connector business logic instead of airbyte protocol knowledge.

Installation Copy Chapter

go get github.com/bitstrapped/airbyte

Docs Copy Chapter

• View the godoc's here: https://pkg.go.dev/github.com/bitstrapped/airbyte

Usage Copy Chapter

By Example Copy Section

1. The fastest way to get started it to look at the full example in examples/httpsource or the Example in the godoc

Detailed Usage Copy Section

1. Define a source by implementing the Source interface.

// Source is the only interface you need to define to create your source!
type Source interface {
	// Spec returns the input "form" spec needed for your source
	Spec(logTracker LogTracker) (*ConnectorSpecification, error)
	// Check verifies the source - usually verify creds/connection etc.
	Check(srcCfgPath string, logTracker LogTracker) error
	// Discover returns the schema of the data you want to sync
	Discover(srcConfigPath string, logTracker LogTracker) (*Catalog, error)
	// Read will read the actual data from your source and use tracker.Record(), tracker.State() and tracker.Log() to sync data with airbyte/destinations
	// MessageTracker is thread-safe and so it is completely find to spin off goroutines to sync your data (just don't forget your waitgroups :))
	// returning an error from this will cancel the sync and returning a nil from this will successfully end the sync
	Read(sourceCfgPath string, prevStatePath string, configuredCat *ConfiguredCatalog,
		tracker MessageTracker) error
}

2. Inside of main, pass your source into the sourcerunner

func main() {
	fsrc := filesource.NewFileSource("foobar.txt")
	runner := airbyte.NewSourceRunner(fsrc)
	err := runner.Start()
	if err != nil {
		log.Fatal(err)
	}
}

3. Write a dockerfile (sample below)

FROM golang:1.17-buster as build

WORKDIR /base
ADD . /base/
RUN go build -o /base/app .


LABEL io.airbyte.version=0.0.1
LABEL io.airbyte.name=airbyte/source

ENTRYPOINT ["/base/app"]

4. Push to your docker repository and profit!

Contributors Copy Section

• We'd like to give a shoutout and thank you to @ajzo90 and his initial work on https://github.com/ajzo90/airbyte-http-connector. @ajzo90's project inspired this project

File: jitsubase/jsoniter/README.md Copy Chapter

A high-performance 100% compatible drop-in replacement of "encoding/json"

Benchmark

Source code: https://github.com/json-iterator/go-benchmark/blob/master/src/github.com/json-iterator/go-benchmark/benchmark_medium_payload_test.go

Raw Result (easyjson requires static code generation)

Always benchmark with your own workload.
The result depends heavily on the data input.

Usage

100% compatibility with standard lib

Replace

import "encoding/json"
json.Marshal(&data)

with

import jsoniter "github.com/json-iterator/go"

var json = jsoniter.ConfigCompatibleWithStandardLibrary
json.Marshal(&data)

Replace

import "encoding/json"
json.Unmarshal(input, &data)

with

import jsoniter "github.com/json-iterator/go"

var json = jsoniter.ConfigCompatibleWithStandardLibrary
json.Unmarshal(input, &data)

More documentation

How to get

go get github.com/json-iterator/go

Contribution Welcomed !

Contributors

• thockin
• mattn
• cch123
• Oleg Shaldybin
• Jason Toffaletti

Report issue or pull request, or email [email protected], or

File: jitsubase/jsonorder/README.md Copy Chapter

A high-performance 100% compatible drop-in replacement of "encoding/json"

Benchmark

Source code: https://github.com/json-iterator/go-benchmark/blob/master/src/github.com/json-iterator/go-benchmark/benchmark_medium_payload_test.go

Raw Result (easyjson requires static code generation)

Always benchmark with your own workload.
The result depends heavily on the data input.

Usage

100% compatibility with standard lib

Replace

import "encoding/json"
json.Marshal(&data)

with

import jsoniter "github.com/json-iterator/go"

var json = jsoniter.ConfigCompatibleWithStandardLibrary
json.Marshal(&data)

Replace

import "encoding/json"
json.Unmarshal(input, &data)

with

import jsoniter "github.com/json-iterator/go"

var json = jsoniter.ConfigCompatibleWithStandardLibrary
json.Unmarshal(input, &data)

More documentation

How to get

go get github.com/json-iterator/go

Contribution Welcomed !

Contributors

• thockin
• mattn
• cch123
• Oleg Shaldybin
• Jason Toffaletti

Report issue or pull request, or email [email protected], or

File: jitsubase/types/README.md Copy Chapter

🔃 github.com/elliotchance/orderedmap/v2 [](https://godoc.org/github.com/elliotchance/orderedmap/v2)

Basic Usage Copy Chapter

An *OrderedMap is a high performance ordered map that maintains amortized O(1)
for Set, Get, Delete and Len:

import "github.com/elliotchance/orderedmap/v2"

func main() {
	m := orderedmap.NewOrderedMap[string, any]()

	m.Set("foo", "bar")
	m.Set("qux", 1.23)
	m.Set("123", true)

	m.Delete("qux")
}

Note: v2 requires Go v1.18 for generics. If you need to support Go 1.17 or
below, you can use v1.

Internally an *OrderedMap uses the composite type
map combined with a
trimmed down linked list to maintain the order.

Iterating Copy Chapter

Be careful using Keys() as it will create a copy of all of the keys so it's
only suitable for a small number of items:

for _, key := range m.Keys() {
	value, _:= m.Get(key)
	fmt.Println(key, value)
}

For larger maps you should use Front() or Back() to iterate per element:

// Iterate through all elements from oldest to newest:
for el := m.Front(); el != nil; el = el.Next() {
    fmt.Println(el.Key, el.Value)
}

// You can also use Back and Prev to iterate in reverse:
for el := m.Back(); el != nil; el = el.Prev() {
    fmt.Println(el.Key, el.Value)
}

The iterator is safe to use bidirectionally, and will return nil once it goes
beyond the first or last item.

If the map is changing while the iteration is in-flight it may produce
unexpected behavior.

File: sync-controller/examples/k8s/README.md Copy Chapter

Example

That is example of entities Sync Controller deploys to k8s cluster to run sync task.
Maybe be outdated and in some details may not reflect current state of entities.
But general idea should be the same.

File: sync-controller/examples/docker-compose/README.md Copy Chapter

Proof of concept

This is proof of concept. Docker-compose runtime is not supported by Sync Controller

File: sync-controller/README.md Copy Chapter

Sync Controller

Sync Controller is a component that is responsible for running sync tasks in kubernetes cluster.
Sync Controller supports source connectors that use Airbyte protocol. All methods are supported: spec, check, discover, read.

Sync Controller run all tasks in separate Pods. It runs pod and monitor its status, updating task status information in database tables.
When task is finished, Sync Controller deletes Pod and frees resources.
K8s Secret used to pass source credentials to Source connector in Pod.

Each Pod contains of two main containers: Source connector and Sync Sidecar.

Sync Sidecar captures data rows, logs, state and results of spec, discover and check command from Source connector:

• Data rows are sent to the target destination in Bulker instance,
• Logs are sent to preconfigured Bulkers destinations.
• spec, discover and check results goes to Postgres database tables.

Authorization Copy Chapter

All request should have Authorization header with value Bearer ${token}.

Token should be one of the tokens defined in SYNCCTL_AUTH_TOKENS env variable,
see configuration section.

If SYNCCTL_AUTH_TOKENS is not configured, then no authorization is required.

Endpoints Copy Chapter

`GET /spec` Copy Section

The endpoint is async. Run task to obtain Source connector specification. Resulting specification is stored in database table source_spec.

Query parameters:

• package* - source package image (for airbyte-based sources it's docker image)
• version* - source package version
• connectorType (optional) - airbyte by default

Result table: source_spec. It writes a last JSON schema with update time by [connectorType, package, version].

It returns either {"status": "ok"} or {"error": ...}

`POST /check` Copy Section

The endpoint is async. Run task to check Source connector credentials.
Check result is stored in database table source_check.

Query parameters:

• storageKey - key to store results of check in a database
• package - source package image (for airbyte-based sources it's docker image)
• version - source package version

Request body:

{
  "config": {
    // credentials json object
  }
}

Result table: source_check

`POST /discover` Copy Section

Run task to obtain Source connector streams catalog.
Catalog is stored in database table source_catalog.

Query parameters:

• storageKey - key to store results of discover in a database
• package - source package image (for airbyte-based sources it's docker image)
• version - source package version

Request body:

{
  "config": {
    // credentials json object
  }
}

Result table: source_catalog

`POST /read` Copy Section

Run task to pull data from Source connector.

• Pulled data is sent to Bulker instance to syncId connection.
• Task logs are inserted into task_log database table.
• Task status is kept updated in source_task database table.
• Saved state of task if any is stored in source_state database table.

Query parameters:

• package - source package image (for airbyte-based sources it's docker image)
• version - source package version
• taskId - id of this task. Should be unique for each task
• syncId - id of sync entity (bulker destination id) where pulled events should be sent

Request body:

{
  "config": {
    // credentials json object
  },
  "catalog": {
    // catalog json object with selected streams and their configurations
  },
  "state": {
    // json object with state from previous task run
  },
  "destinationConfig": {
    // bulker destination credentials
  }
}

Configuration Copy Chapter

Database schema Copy Chapter

File: sync-sidecar/README.md Copy Chapter

🏍️ Sync Sidecar

Sync Sidecar is meant to be used as sidecar container to Airbyte protocol compatible Source connectors in Kubernetes Pod or Docker Compose.

Sync Sidecar captures data rows, logs, state and results of spec, discover and check command from Source connector:

• Data rows are sent to the target destination in Bulker instance,
• Logs are sent to preconfigured Bulkers destinations.
• spec, discover and check results goes to Postgres database tables.

Named Pipes Copy Chapter

Bulker-Sidecar uses named pipes to communicate with Source connector.
In k8s environment all containers from the same Pod are running on the same machine, and it is possible to use named pipes.
Source connectors entrypoint must be changed to direct their stderr and stdout output to named pipes.

Volume with named pipes should be mounted to both Bulker-Sidecar and Source connector containers.
InitContainer can be used to create named pipes in advance.

Configuration Copy Chapter

• STDOUT_PIPE_FILE - path of named pipe for stdout of Source connector
• STDERR_PIPE_FILE - path of named pipe for stderr of Source connector
• COMMAND - Command that is used to run Source connector. Should be one of spec, discover, check or read
• STARTED_AT - Timestamp when task was triggered
• DATABASE_URL - URL of Postgres database where spec, discover, check results and read task statuses should be stored
• PACKAGE - Name of Source connector package
• PACKAGE_VERSION - Version of Source connector package
• STORAGE_KEY - key to store results of check and discover commands in a database.
• SYNC_ID - id of sync entity (bulker destination id) where pulled events should be sent. For read command.
• TASK_ID - id of current running task

2. Official Technical Reference & Guides (jitsucom/docs) Copy Chapter

File: README.md Copy Chapter

Introduction

kSense helps businesses collect, correlate and perform action based on customer intentions.

File: events-api/README.md Copy Chapter

Events API

Introduction Copy Section

kSense provides event API for exporting events to its analytics and prediction system. Events can have different meaning (depending on purpose of client integration). It could be web page views, mobile app usage, conversions, subscription / un-subscription events and so on. Moreover, user can define a custom event types.

There're two ways of exporting events: through JavaScript call (pixel) and server-to-server. Both options have the same JSON API with minor differences in user-indentification process. All definitions are applicable to both JavaScript and JSON APIs.

Basic concepts Copy Section

Event Type

Each event has a type name. Some event types are pre-defined (system even types). Also, user can define its own type name. Each event may have a parent event type. It means that event can happen if only a parent type has happened before for the same user. System event types:

Please see Event JSON page for further documentation.

File: case-scenarios/README.md Copy Chapter

Case Scenarios

This section describes how event API should be used for different type of consumers: which event to send and what is the impact.