Topic

Topic is the core concept of Watchmen,

• Is categorized by kind and type,
• Has a set of factors,
• Attached into a data source
  • In RDS, topic equates table,
  • In NoSQL, topic equates document.

And

• All data are stored in topics,
• Pipelines are run on topics,
• Subjects, datasets, reports are built based on topics.

info

Topic is available for Admin only.

Standard Tuple Page

Topic use standard tuple page.

Kind

There are 2 kinds of topic,

• System
• Business

For example, topics for business data are Business, topics for log data are System.

Type

There are several types of topic.

Raw

Raw topic is for receiving data from other systems, we use JSON format to store raw data which received via RESTful API. Raw data is immutable, which means each request will create a new raw data record.
In some cases, raw data needs to be searched, but as a fairly well known fact, data stored as JSON is not that easy for searching, especially in RDS, such as Oracle and MySQL. In the latest version of Oracle and MySQL, they both provide the functions to operate JSON objects, but we offer an easy way, which is factor in hierarchical structure of JSON object can be flattened as a single column.

For example, here is an order JSON,

{
    "orderId": 10000,
    "orderNo": "N10000",
    "amount": 10000,
    "customer": {
        "name": "John Doe",
        "gender": "M"
    }
}

Simply tick the Flatten Column checkbox, a single table column is created for customer.name,

Now, the raw topic can be searched by simple SQL,

// assume topic name is "customer", then table name will be "TOPIC_CUSTOMER"
SELECT * FROM TOPIC_CUSTOMER WHERE CUSTOMER_NAME = 'John Doe';

info

• Index doesn't be applied in above sample. In real world, for search performance, it should be declared,
• Dots(.) in factor name are replaced by underscore.

Meta

Meta topic is for those data is definition, meta, not for instance data. There are several characteristic features of metadata,

• Includes more factors instead of code-label pair in enumeration,
• Immutable or at least slowly changing,
• No writing from other topics,
• Reading by pipelines only.

Subject/dataset will be more descriptive base on a set of well-designed meta topics.

Distinct

Distinct topic is for instance data, such as order, customer, etc. In transaction system, instance data is changed time by time, every snapshot should be extracted and send to Watchmen, store in distinct topic. On the other hand, we also know change logs are very important part in IT system, for some reason, typically for audit, changes are logged and persisted. From perspective of changes on timeline, they are also can be treated as distinct data. Let's play that out with an example to make it more apparent.

• Firstly, a raw topic RAW_ORDER,

{
    "orderId": 10000,
    "orderNo": "N10000",
    "amount": 10000,
    "customer": {
        "name": "John Doe",
        "gender": "M"
    }
}

• Create a distinct topic ORDER for instance data, a distinct row will be created when raw data received,

OrderID	OrderNO	Amount
10000	N10000	10000

• Create another distinct topic ORDER_AMOUNT_CHANGES for change logs of order amount, a change row will be created,

ChangeID	OrderID	OrderNO	Amount
9000000000000000000	10000	N10000	10000

• Change occurred on order N10000, amount is changed from 10000 to 20000,

{
    "orderId": 10000,
    "orderNo": "N10000",
    "amount": 20000,
    "customer": {
        "name": "John Doe",
        "gender": "M"
    }
}

• For topic ORDER, change should be merged into order record,

OrderID	OrderNO	Amount
10000	N10000	20000

• For topic ORDER_AMOUNT_CHANGES, a new record is created for change,

ChangeID	OrderID	OrderNO	Amount
9000000000000000000	10000	N10000	10000
9000000000000000001	10000	N10000	20000

Now we see the difference between topics ORDER and ORDER_AMOUNT_CHANGES, they describe one order from different perspectives. From statistical experience, different perspectives will lead the different conclusions. In this case, we can analyze the consumer behavior of how much from ORDER, and how often changes from ORDER_AMOUNT_CHANGES.

info

Values of ChangeID are generated by built-in sequence generating function, find it on pipeline.

Aggregation

Watchmen offers 3 types of topics to store aggregation data.

• Aggregate
• Time
• Ratio

Types of aggregation topics are designed from the point of view of management. It is suggested to classify aggregation topics to,

• Ratio: has one computed ratio factor at least, for aggregated to a ratio, like increment, decrement, etc.,
• Time: has one time-dimension factor at least, for aggregation on time period, like annually, quarterly, monthly, etc.,
• Aggregate: except ratio and time aggregation, for aggregation on enumeration, meta, etc.

Factor

Factors are the basic elements of a topic. According to topic stored, factor equates

• A table column when topic is a table in RDS,
• A field when topic is a document in NoSQL.

Factor Types

Watchmen offers 50+ types for factor,

Factor Type	Comment	Purpose
Sequence	Sequence, auto-increment
Number	Numeric value
Unsigned	0 & Positive numeric value
Text	String value
Boolean	Boolean value
Address	String value for address	Geo related
Continent	Predefined enumeration, for address	Geo related
Region	Predefined enumeration, for address	Geo related
Country	Predefined enumeration, for address	Geo related
Province	Predefined enumeration, for address	Geo related
City	Predefined enumeration, for address	Geo related
District	Predefined enumeration, for address	Geo related
Road	Predefined enumeration, for address	Geo related
Community	String value
Floor	Numeric value for address
Residence type	Predefined enumeration
Residential area	Numeric value for house property
Email	String value for contact
Phone	String value for contact
Mobile	String value for contact
Fax	String value for contact
Datetime	YYYY-MM-DD HH:mm:ss	Time related
Full datetime	YYYY-MM-DD HH:mm:ss.SSS	Time related
Date	YYYY-MM-DD	Time related
Time	HH:mm:ss	Time related
Year	4 digits	Time related
Half year	1: first half, 2: second half	Time related
Quarter	1 - 4	Time related
Month	1 - 12	Time related
Half month	1: first half, 2: second half	Time related
Ten days	1, 2, 3	Time related
Week of year	0 (the partial week that precedes the first Sunday of the year) - 53 (leap year)	Time related
Week of month	0 (the partial week that precedes the first Sunday of the year) - 5	Time related
Half week	1: first half, 2: second half	Time related
Day of month	1 - 31, according to month/year	Time related
Day of week	1 (Sunday) - 7 (Saturday)	Time related
Day kind	1: workday, 2: weekend, 3: holiday	Time related
Hour	0 - 23	Time related
Hour kind	1: work time, 2: off hours, 3: sleeping time	Time related
Minute	0 - 59	Time related
Second	0 - 59	Time related
Millisecond	0 - 999	Time related
Am pm	1, 2	Time related
Gender	Predefined enumeration	Individual related
Occupation	Predefined enumeration	Individual related
Date of birth	YYYY-MM-DD	Individual related
Age	Positive numeric value	Individual related
ID No.		Individual related
Religion	Predefined enumeration	Individual related
Nationality	Predefined enumeration	Individual related
Biz trade	Predefined enumeration	Organization related
Biz scale	Positive numeric value	Organization related
Enum	Enumerated values, should be defined with Enumeration

caution

We highly recommend defining proper type for factor. Type is more readable than name, it can be used to do something in further analysis.

Special Types for Raw Topic

There are 2 types for raw topic only,

Factor Type	Comment	Purpose
Object		Raw topic only
Array		Raw topic only

Because of raw topic store data exactly same as what RESTful received, which is on JSON format, therefore hierarchical structure is necessary. These 2 types are designed to describe the object and array element. Factors in object or array element need to be declared explicitly, use dots(.) to define the hierarchy.

For example, a raw topic instance data as below,

{
    "orderId": 10000,
    "orderNo": "N10000",
    "amount": 20000,
    "customer": {
        "name": "John Doe",
        "gender": "M",
        "contact": {
            "mobile": "87654321",
            "fax": "12345678"
        },
        "addresses": [
            {
                "city": "New York",
                "district": "Brooklyn",
                "line1": "26 W. Talbot Avenue",
                "line2": ""
            },
            {
                "city": "New York",
                "district": "South Richmond Hill",
                "line1": "7727 Stonybrook St."
            }
        ]
    }
}

Factors should be defined as below,

Name	Type	Flatten	Index
orderId	Sequence	✅	✅
orderNo	Text	✅	✅
amount	Unsigned
customer	Object
customer.name	Text	✅	✅
customer.gender	Gender
customer.contact	Object
customer.contact.mobile	Mobile	✅	✅
customer.contact.fax	Fax
customer.addresses	Array
customer.addresses.city	City
customer.addresses.district	District
customer.addresses.line1	Text
customer.addresses.line2	Text

info

Typically, a factor should be indexed when it is flattened.

caution

Factors belong to an array cannot be flattened and indexed.

Encryption

For security reason, factor can be encrypted. There are several encryption methods built-in,

• AES256 PKCS5 PADDING: reversible,
• MD5: one-way,
• SHA256: one-way,
• MASK MAIL: mask mail,
• MASK CENTER 3: mask 3 characters in the middle,
• MASK CENTER 5: mask 5 characters in the middle,
• MASK LAST 3: mask 3 characters at the end,
• MASK LAST 6: mask 6 characters at the end,
• MASK DAY: mask day in time type,
• MASK MONTH: mask month in time type,
• MASK MONTH DAY: mask month and day in time type.

Default Value

Default value can be set on factor.

caution

Type of default value will not be validated, keep it compatible with factor type.

Index & Unique Index

In raw topic, flattened factor should be indexed for searching purpose. For other types of topics, if factor will be used in pipelines as criteria, and topic is stored in RDS, never forget the index, it improves query performance by avoiding a full table scan.

Watchmen offers 10 index groups per topic, factors sharing the same index group will be created as a composite index. Use unique index group instead when instance data of factor are unique, Watchmen also offers 10 unique index groups per topic. How to group index depends on the query criteria in pipelines, it is similar with RDS index.

caution

Sequence type might be not unique in topic, such as a orderId in ORDER_AMOUNT_CHANGES, it is a Sequence, but not unique.

Enumeration

When a factor is defined as a Enum, relationship with Enumeration should be defined as well, system know the range of values by this. It is important for UI rendering, and also is important for monitoring in DQC.

info

See Enumeration and DQC for more information.

Label & Description

Label is for human reading on UI rendering, description is for more details on system maintaining.

Factor Operations

Search

A search bar is on above, factors will be filtered according to searching text.

• A text: try to match name and label of factor,
• n:name: match name only, such as n:order,
• l:label: match label only, such as l:seq,
• t:type: match type only, such as t:text,
• i:index: match index only, such as i:1,
• u:index: match unique index only, such as u:1,
• v:value: match default value only,
  • List factors which have default value when v:true,
  • List factors which have no default value when it is not v:true, for short v: also works,
• e:enum: match enumerated factors only,
  • List factors which have enumeration when e:true,
  • List factors which have no enumeration when it is not e:true, for short e: also works,
  • Only effective on enumeration factors,
• f:flatten: match flattened factors only,
  • List factors which are flattened when f:true,
  • List factors which are not flattened when it is not f:true, for short f: also works,
  • Only effective on raw topic,
• c:crypt: match encrypted factors only,
  • List factors which are encrypted when c:true,
  • List factors which are not encrypted when it is not c:true, for short c: also works.

Pagination

Factor list is pagination, there are 50 factors per page, switching by pagination dropdown.

info

Pagination is under filter.

Add & Delete

Factors are sorted exactly by where they were added, normally it is not a problem, but in sometimes, new factor would be inserted before existing. Use Prepend Factor button in factor buttons to insert factor, or use Append Factor button in factors buttons to append factor at the end.

info

Page will be switched automatically when the appended factor is in next page.

caution

DO NOT prepend/append factor when a filter is applied, new factor might be invisible because it is filtered by search text.

Delete factor by Delete Factor in factor buttons.

caution

Deletion cannot be recovered, must be re-added again. Be careful on this.

Import

Factors can be imported by file, there are several ways to import factors,

• By structure file,
  • CSV, or TXT in csv format,
  • JSON,
• By instance data file,
  • JSON.

caution

Factors will be replaced by imported.

Structure

A structure template zip can be downloaded by Download Structure Template button.

Files in zip as below,

zip root/
├── factor-template.csv
└── factor-template.json

Modify the template file simply following guide inside, and import by Import By Structure button.

Instance Data

For easy to import, instance data file is also supported. A json file including data can be imported and factors are detected automatically. Make sure data in file is an object array, including the sampling data items. Importer detects factors by following rules:

• First type is better. Using the first detected type, or raise conflict exception,
  • Text is compatible with any types excepts Object/Array,
  • All null values should be detected as Text,
• Datetime type first for text value,
  • 23 digits for Full Datetime,
  • 19 digits for DateTime,
  • 10 digits for Date,
  • 8 digits for Time,
• Object/Array is only for raw topic, let topic be raw type before importing.

For example,

[
    {
        "orderId": 10000,                                   // number detected
        "orderNo": "N10000",                                // text detected
        "amount": null,                                     // null value, suspended for now,
        "enabled": true,                                    // boolean detected
        "customer": {                                       // object detected
            "name": "John Doe",
            "gender": "M",                                  // text detected, not gender
            "contact": {
                "mobile": "87654321",
                "fax": "12345678"
            },
            "addresses": [                                  // array detected
                {
                    "city": "New York",
                    "district": "Brooklyn",
                    "line1": "26 W. Talbot Avenue",
                    "line2": ""
                },
                null,                                       // ignored when not an object in array
                10000                                       // ignored when not an object in array
            ]
        }
    },
    {
        "orderId": "10001",                                 // incorrect, detected as number in first item
        "orderNo": 10001,                                   // correct, text is compatible with number
        "amount": null,                                     // all values are null, detected as text     
        "enabled": null,                                    // null, is compatible with boolean
    }
]

Scripts

Check supported script types by following table,

Script Purpose	Oracle SQL Script	Oracle Liquibase	MySQL SQL Script	MySQL Liquibase
Creation	✅	✅	✅	✅
Alteration	✅	✅	✅	✅

How to Export Scripts

• Click Download Scripts button,
• Pick topics which need to be exported,
• Pick database types,
• Pick script purposes,
• Pick script formats,
• Click Download button.

A zip file is exported. If all checkbox are ticked, folders should be as below,

zip root/
├── liquibase
│   └── order.xml
├── oracle
│   ├── alteration
│   │   └── order.sql
│   └── creation
│       └── order.sql
└── mysql
    ├── alteration
    │   └── order.sql
    └── creation
        └── order.sql

• Liquibase scripts are combined in one file,
• SQL scripts are separated in different folders.

How Script Works

For Creation

• Table creation will be ignored on liquibase script, but this will not be checked on SQL script,
• SQL for drop table is commented in script file, uncomment it if you need,
  Note drop table will drop all instance data also,
• Mandatory columns will be created automatically,
• PK column: id_,
• Aggregation column: aggregate_assist_,
• Version column: version_,
• Data zone column: tenant_id_,
• Time audit columns: insert_time_ and update_time_,
• Raw data column: data_, available for raw topic only,
• Primary key will be created automatically on PK column,
• Indexes will be created automatically,
• On data zone column,
• On time audit columns,
• On indexed factors,
• Unique indexes on factors which appointed,
• this part is shared between liquibase creation and alteration.

For Alteration

Alteration is very similar with creation, except columns will be

• Added when not existed,
• Modified when existed.

tip

Read comments in script files for more details about how to choose, modify and execute them.

Be very careful on execution of scripts.

Scripts exported are base version generated by frontend, we have no idea with what exactly is the structure of target data source.
Therefore, we highly recommend review all script files, modify them to what you need, store and maintain versions by some version control tool, such as GIT.

Form Validation

• Topic Name: required,
• Must follow the standard of data source type. For example, in oracle, topic equates a Table,
• Snake case or camel case is allowed,
• Topic Kind: required,
• Topic Type: required,
• Data Source: required,
• Description: optional,
• Factors: at least one factor is required,
• Name: required,
• Label: optional. Highly recommend filling it with a human reading text,
• Type: required,
• Default Value: optional,
• Index Group: optional,
• Encryption: optional,
• Enumeration: required when a factor is an enumeration,
• Flatten Column: required when topic is raw and factor name contains at least one dot(.).

Topic Layers

There are a large amount of topics should be defined, we recommend that topics would be defined on different layers. Each layer has its own purpose, and using pipelines lead data through these layers, cleaning, transforming and aggregating them.