th2-data-services 2.0.0.dev6022917985

th2_data_services

Table of Contents

• Table of Contents
• 1. Introduction
• 2. Getting started
  • 2.1. Installation
    • Core
    • Data sources (providers)
  • 2.2. Example
  • 2.3. Short theory
    • Terms
    • Concept
    • Stream operations
      • Pipelining
      • Internal iteration
    • Data caching
      • Forced caching
    • EventTree and collections
      • EventTree
      • Collections
      • Hints
    • FieldsResolver
  • 2.4. Links
• 3. Best practices
• 4. Official DataSource implementations
• 5. API
• 6. Examples

1. Introduction

This repository is a library for creating th2-data-services applications. Data Services allows you to manipulate the stream data processing workflow using pipelining.

The library's features:

• Provides core interface for developing data source implementations
• Work with iterable objects (list, tuple, etc including files) via Data object using its features
• Manipulate the workflow to make some analysis by Data object methods
• Use timestamp converter implementations or use base class to create custom converters
• Build Event Trees (EventTree, EventTreeCollection and ParentEventTreeCollection classes)

Workflow manipulation tools allows you:

• Filtering stream data (Data.filter method)
• Transforming stream data (Data.map method)
• Limiting the number of processed streaming data (Data.limit method)

There is also another part of data services

• th2-data-services-utils. It's a set of tools to perform the most common analysis tasks.

2. Getting started

2.1. Installation

Core

• From PyPI (pip)
  This package can be found on PyPI.

  pip install th2-data-services
  

• From Source

  git clone https://github.com/th2-net/th2-data-services
  pip install th2-data-services/
  

Data sources (providers)

Since v1.3.0, the library doesn't provide data source dependencies.

You should provide it manually during installation. You just need to add square brackets after library name and put dependency name.

pip install th2-data-services[dependency_name]

Dependencies list

dependency name	provider version
lwdp	latest version of lwdp
lwdp2	latest version of lwdp v2
utils-rpt-viewer	latest version of utils-rpt-viewer
utils-rpt-viewer5	latest version of utils-rpt-viewer v5
utils-advanced	latest version of ds-utils

Example

pip install th2-data-services[lwdp1]

2.2. Example

A good, short example is worth a thousand words.

This example shows basic usage of library's features.

The following example as a file.

from typing import Tuple, List, Optional, Generator
from datetime import datetime

from th2_data_services.data import Data
from th2_data_services.event_tree import EventTree, EventTreeCollection, ParentEventTreeCollection, IETCDriver
from th2_data_services.interfaces import IDataSource
from th2_data_services.utils.converters import DatetimeConverter, DatetimeStringConverter, ProtobufTimestampConverter

# [0] Lib configuration
# [0.1] Interactive or Script mode
# If you use the lib in interactive mode (jupyter, ipython) it's recommended to set the special
# global parameter to True. It'll keep cache files if something went wrong.
from th2_data_services.config import options

options.INTERACTIVE_MODE = True

# Some example data
events = Data(
    [
        {
            "eventId": "demo_book_1:th2-scope:20230105135705560873000:d61e930a-8d00-11ed-aa1a-d34a6155152d_1",
            "batchId": None,
            "isBatched": False,
            "eventName": "Set of auto-generated events for ds lib testing",
            "eventType": "ds-lib-test-event",
            "endTimestamp": {"epochSecond": 1672927025, "nano": 561751000},
            "startTimestamp": {"epochSecond": 1672927025, "nano": 560873000},
            "parentEventId": None,
            "successful": True,
            "bookId": "demo_book_1",
            "scope": "th2-scope",
            "attachedMessageIds": [],
            "body": [],
        },
        {
            "eventId": "demo_book_1:th2-scope:20230105135705563522000:9adbb3e0-5f8b-4c28-a2ac-7361e8fa704c>demo_book_1:th2-scope:20230105135705563522000:d61e930a-8d00-11ed-aa1a-d34a6155152d_2",
            "batchId": "demo_book_1:th2-scope:20230105135705563522000:9adbb3e0-5f8b-4c28-a2ac-7361e8fa704c",
            "isBatched": True,
            "eventName": "Plain event 1",
            "eventType": "ds-lib-test-event",
            "endTimestamp": {"epochSecond": 1672927025, "nano": 563640000},
            "startTimestamp": {"epochSecond": 1672927025, "nano": 563522000},
            "parentEventId": "demo_book_1:th2-scope:20230105135705560873000:d61e930a-8d00-11ed-aa1a-d34a6155152d_1",
            "successful": True,
            "bookId": "demo_book_1",
            "scope": "th2-scope",
            "attachedMessageIds": [],
            "body": {"type": "message", "data": "ds-lib test body"},
        },
        {
            "eventId": "demo_book_1:th2-scope:20230105135705563522000:9adbb3e0-5f8b-4c28-a2ac-7361e8fa704c>demo_book_1:th2-scope:20230105135705563757000:d61e930a-8d00-11ed-aa1a-d34a6155152d_3",
            "batchId": "demo_book_1:th2-scope:20230105135705563522000:9adbb3e0-5f8b-4c28-a2ac-7361e8fa704c",
            "isBatched": True,
            "eventName": "Plain event 2",
            "eventType": "ds-lib-test-event",
            "endTimestamp": {"epochSecond": 1672927025, "nano": 563791000},
            "startTimestamp": {"epochSecond": 1672927025, "nano": 563757000},
            "parentEventId": "demo_book_1:th2-scope:20230105135705560873000:d61e930a-8d00-11ed-aa1a-d34a6155152d_1",
            "successful": True,
            "bookId": "demo_book_1",
            "scope": "th2-scope",
            "attachedMessageIds": [],
            "body": {"type": "message", "data": "ds-lib test body"},
        },
    ]
)

# [1] Working with a Data object.
# [1.1] Filter.
filtered_events: Data = events.filter(lambda e: e["body"] != [])  # Filter events with empty body.


# [1.2] Map.
def transform_function(record):
    return {"eventName": record["eventName"], "successful": record["successful"]}


filtered_and_mapped_events = filtered_events.map(transform_function)

# [1.3] Data pipeline.
#       Instead of doing data transformations step by step you can do it in one line.
filtered_and_mapped_events_by_pipeline = events.filter(lambda e: e["body"] != []).map(transform_function)
# Content of these two Data objects should be equal.
assert list(filtered_and_mapped_events) == list(filtered_and_mapped_events_by_pipeline)

# [1.4] Sift. Skip the first few items or limit them.
data = Data([1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15])
items_from_11_to_end: Generator = data.sift(skip=10)
only_first_10_items: Generator = data.sift(limit=10)

# [1.5] Changing cache status.
events.use_cache(True)
# or just
events.use_cache()  # If you want to activate cache.
# [1.6] Walk through data.
for event in events:
    # Do something with event (event is a dict).
    print(event)
# After first iteration the events has a cache file.
# Now they will be used in the cache in the next iteration.

# [1.7] Get number of the elements in the Data object.
number_of_events = events.len

# [1.8] Check that Data object isn't empty.
# The data source should be not empty.
assert events.is_empty is False

# [1.9] Convert Data object to the list of elements(events or messages).
# Be careful, this can take too much memory.
events_list = list(events)

# [1.10] The cache inheritance.
# Creates a new Data object that will use cache from the events Data object.
events_filtered: Data = events.filter(lambda record: record.get("batchId"))

# New Data objects don't use their own cache by default but use the cache of the parent Data object.
# Use use_cache method to activate caching.
# After that, the Data object will create its own cache file.
events_filtered.use_cache()

list(events_filtered)  # Just to iterate Data object (cache file will be created).

filtered_events_types = events_filtered.map(lambda record: {"eventType": record.get("eventType")})

events_without_types_with_batch = filtered_events_types.filter(lambda record: not record.get("eventType"))
events_without_types_with_batch.use_cache()

# [1.11] Data objects joining.
# You have the following 3 Data objects.
d1 = Data([1, 2, 3])
d2 = Data(["a", {"id": 123}, "c"])
d3 = Data([7, 8, 9])
# You can join Data objects in following ways.
# Please note, new Data object will have cache status == False.
data_via_init = Data([d1, d2, d3])
data_via_add = d1 + d2 + d3
data_with_non_data_obj_via_init = Data([d1, ["a", {"id": 123}, "c"], d3])
data_with_non_data_obj_via_add = d1 + ["a", {"id": 123}, "c"] + d3
# You can join current Data object on place using +=.
# It will keep cache status.
d1 += d3  # d1 will become Data([1,2,3,7,8,9])

# [1.12] Build and read Data object cache files.
events.build_cache("cache_filename_or_path")
data_obj_from_cache = Data.from_cache_file("cache_filename_or_path")

# [2] Working with converters.
# There are currently three implementations of ITimestampConverter class: DatetimeConverte, DatetimeStringConverter and ProtobufTimestampConverter.
# They all implement same methods from base class.
# Note that some accuracy may be lost during conversion.
# If for example you use to_microseconds nanoseconds will be cut off instead of rounding.

# [2.1] DatetimeConverter.
# DatetimeConverter takes datetime.datetime object as input.

datetime_obj = datetime(year=2023, month=1, day=5, hour=14, minute=38, second=25, microsecond=1460)

# It has methods that return the datetime in different formas:

date_ms = DatetimeConverter.to_milliseconds(datetime_obj)
date_us = DatetimeConverter.to_microseconds(datetime_obj)
# Converting to nanoseconds justs adds three trailing zeros as datetime object doesn't have nanoseconds.
date_ns = DatetimeConverter.to_nanoseconds(datetime_obj)

# [2.2] DatetimeStringConverter
# DatetimeStringConverter takes string in "yyyy-MM-ddTHH:mm:ss[.SSSSSSSSS]Z" format.

date_string = "2023-01-05T14:38:25.00146Z"

# We have same methods as in DatetimeConverter
date_ms_from_string = DatetimeStringConverter.to_milliseconds(date_string)
date_us_from_string = DatetimeStringConverter.to_microseconds(date_string)
date_ns_from_string = DatetimeStringConverter.to_nanoseconds(date_string)

# We can also get datetime object from string
datetime_from_string = DatetimeStringConverter.to_datetime(date_string)

# [2.3] ProtobufTimestampConverter
# Protobuf timestamps must be in form {"epochSecond": seconds, "nano": nanoseconds}

protobuf_timestamp = {"epochSecond": 1672929505, "nano": 1_460_000}

date_ms_from_timestamp = ProtobufTimestampConverter.to_milliseconds(protobuf_timestamp)
date_us_from_timestamp = ProtobufTimestampConverter.to_microseconds(protobuf_timestamp)
date_ns_from_timestamp = ProtobufTimestampConverter.to_nanoseconds(protobuf_timestamp)
datetime_from_timestamp = ProtobufTimestampConverter.to_datetime(protobuf_timestamp)

# [3] Working with EventTree and EventTreeCollection.

# [3.1] Build a custom EventTree
# To create an EventTree object you need to provide name, id and data of the root event.
tree = EventTree(event_name="root event", event_id="root_id", data={"data": [1, 2, 3, 4, 5]})

# To add new node use append_event. parent_id is necessary, data is optional.
tree.append_event(event_name="A", event_id="A_id", data=None, parent_id="root_id")

# [3.3] Building the EventTreeCollection.

# If you don't specify data_source for the driver then it won't recover detached events.
driver: IETCDriver  # You should init ETCDriver object. E.g. from LwDP module or your custom class.
etc = EventTreeCollection(driver)
etc.build(events)

# Detached events isn't empty.
assert etc.get_detached_events()

etc = EventTreeCollection(driver)
# Detached events are empty because they were recovered.
assert not etc.get_detached_events()

# The collection has EventTrees each with a tree of events.
# Using Collection and EventTrees, you can work flexibly with events.

# [3.3.1] Get leaves of all trees.
leaves: Tuple[dict] = etc.get_leaves()

# [3.3.2] Get roots ids of all trees.
roots: List[str] = etc.get_roots_ids()

# [3.3.3] Find an event in all trees.
find_event: Optional[dict] = etc.find(lambda event: "Send message" in event["eventType"])

# [3.3.4] Find all events in all trees. There is also iterable version 'findall_iter'.
find_events: List[dict] = etc.findall(lambda event: event["successful"] is True)

# [3.3.5] Find an ancestor of the event.
ancestor: Optional[dict] = etc.find_ancestor(
    "8bbe3717-cf59-11eb-a3f7-094f904c3a62", filter=lambda event: "RootEvent" in event["eventName"]
)

# [3.3.6] Get children of the event. There is also iterable version 'get_children_iter'.
children: Tuple[dict] = etc.get_children("814422e1-9c68-11eb-8598-691ebd7f413d")

# [3.3.7] Get subtree for specified event.
subtree: EventTree = etc.get_subtree("8e23774d-cf59-11eb-a6e3-55bfdb2b3f21")

# [3.3.8] Get full path to the event.
# Looks like [ancestor_root, ancestor_level1, ancestor_level2, event]
event_path: List[dict] = etc.get_full_path("8e2524fa-cf59-11eb-a3f7-094f904c3a62")

# [3.3.9] Get parent of the event.
parent = etc.get_parent("8e2524fa-cf59-11eb-a3f7-094f904c3a62")

# [3.3.10] Append new event to the collection.
etc.append_event(
    event={
        "eventId": "a20f5ef4-c3fe-bb10-a29c-dd3d784909eb",
        "parentEventId": "8e2524fa-cf59-11eb-a3f7-094f904c3a62",
        "eventName": "StubEvent",
    }
)

# [3.3.11] Show the entire collection.
etc.show()

# [3.4] Working with the EventTree.
# EventTree has the same methods as EventTreeCollection, but only for its own tree.

# [3.4.1] Get collection trees.
trees: List[EventTree] = etc.get_trees()
tree: EventTree = trees[0]

# But EventTree provides a work with the tree, but does not modify it.
# If you want to modify the tree, use EventTreeCollections.

# [3.5] Working with ParentlessTree.
# ParentlessTree is EventTree which has detached events with stubs.
parentless_trees: List[EventTree] = etc.get_parentless_trees()

# [3.6] Working with ParentEventTreeCollection.
# ParentEventTreeCollection is a tree collection like EventTreeCollection,
# but it has only events that have references.
data_source: IDataSource  # You should init DataSource object. E.g. from LwDP module.
# ETCDriver here is a stub, actually the lib don't have such class.
# You can take it in LwDP module or create yourself class if you have some special events structure.
driver = ETCDriver(data_source=data_source)
etc = ParentEventTreeCollection(driver)
etc.build(events)

etc.show()

2.3. Short theory

The library provides tools for handling stream data. What's a stream? It's a sequence of elements from a source that supports aggregate operations.

Terms

• Data object: An instance of Data class which is wrapper under stream.
• Sequence of elements: A Data object provides an interface to a sequenced set of values of a specific element type. Stream inside the Data object dont actually store elements; they are computed on demand.
• data source (exactly in small letters): Any source of data. E.g. Lightweight Data Provider, collections, arrays, or I/O resources.
• DataSource: A class that is an intermediate link between the SourceAPI and Commands.
• SourceAPI: Each source has its own API to retrieve data. SourceAPI is a class that provide API for some data source.
• Commands: Classes that provide user-friendly interfaces for getting some data from DataSource. Commands use SourceAPI to achieve it.
• Adapters: It's similar to function for Data.map method. Adoptable commands used it to update the data stream.
• Aggregate operations: Common operations such as filter, map, limit and so on.
• Workflow: An ordered set of Aggregate operations.

Concept

The library describes the high-level interfaces ISourceAPI, IDataSource, ICommand, IAdapter.

Any data source must be described by the IDataSource abstract class. These can be FileDataSource, CSVDataSource, DBDataSource and other.

Usually, data sources have some kind of API. Databases - provide SQL language, when working with a file, you can read line by line, etc. This API is described by the ISourceAPI class. Because different versions of the same data source may have different API, it is better to create a class for each version.

Generally, data source APIs are hidden behind convenient interfaces. The role of these interfaces is played by ICommand classes.

IAdapter classes transform data stream like functions for Data.map method. Essentially it's the same thing but more flexible.

For example, LwDP DataSource(https://github.com/th2-net/th2-ds-source-lwdp) uses these abstract classes to build its implementation.You can easily create your own unique commands for LwDP DataSource, as well as entire DataSource classes. Here is a documentation on how to implement these interfaces.

Stream operations

Furthermore, stream operations have two fundamental characteristics that make them very different from collection operations: Pipelining and Internal iteration.

Pipelining

Many stream operations return a stream themselves. This allows operations to be chained to form a larger pipeline.

Internal iteration

In contrast to collections, which are iterated explicitly (external iteration), stream operations do the iteration behind the scenes for you. Note, it doesn't mean you cannot iterate the Data object.

Data caching

The Data object provides the ability to use the cache. The cache works for each Data object, that is, you choose which Data object you want to save. The Data object cache is saved after the first iteration, but the iteration source may be different.

If you don't use the cache, your source will be the data source you have in the Data Object. But if you use the cache, your source can be the data source, the parent cache, or own cache:

• The data source: If the Data Object doesn't have a parent cache and its cache.
• The parent cache: If the Data Object has a parent cache. It doesn't matter what position the parent cache has in inheritance. Data Object understands whose cache it is and executes the part of the workflow that was not executed.
• The own cache: If it is not the first iteration of this Data object.

Note that the cache state of the Data object is not inherited.

Forced caching

You can tell DS to cache data to specific cache file, which won't be deleted after script end. You can see example in 1.12 section of get_started_example.

EventTree and collections

EventTree

EventTree is a tree-based data structure of events. It allows you get children and parents of event, display tree, get full path to event etc.

Details:

• EventTree contains all events in memory.
• Tree has some important terms:
  1. Ancestor is any relative of the event up the tree (grandparent, parent etc.).
  2. Parent is only the first relative of the event up the tree.
  3. Child is the first relative of the event down the tree.

Take a look at the following HTML tree to understand them.

 <body> <!-- ancestor (grandparent), but not parent -->
     <div> <!-- parent & ancestor -->
         <p>Hello, world!</p> <!-- child -->
         <p>Goodbye!</p> <!-- sibling -->
     </div>
 </body>

Collections

EventTreeCollection is a collection of EventTrees. The collection builds a few EventTree by passed Data object. Although you can change the tree directly, it's better to do it through collections because they are aware of detached_events and can solve some events dependencies. The collection has similar features like a single EventTree but applying them for all EventTrees.

ParentEventTreeCollection is a collection similar to EventTreeCollection but containing only parent events that are referenced in the data stream. The collection has features similar to EventTreeCollection.

Details:

• To use ET collections you need to initialize them by ETCDriver. Data sources usually provide them. You can create it by yourself depending on your data structure.
• The collection has a feature to recover events. All events that are not in the received data stream, but which are referenced will be loaded from the data source.
• You can take detached_events to see which events are missing.
• If you want, you can build parentless trees where the missing events are stubbed instead. Just use get_parentless_trees().

Requirements:

1. Events provided to ETC have to have event_name, event_id, parent_event_id fields. They can have another names (it resolves in the driver).

Hints

• Remove all unnecessary fields from events before passing to a collection to reduce memory usage.
• Use show() method to print the tree in tree-like view.
• Note that the get_x methods will raise an exception if you pass an unknown event id, unlike the find_x methods ( they return None).
• If you want to know that specified event exists, use the python in keyword (e.g. 'event-id' in events_tree).
• Use the python len keyword to get events number in the tree.

FieldsResolver

The idea of using resolvers:
It solves the problem of having a few DataSources with the same data, but with different ways to get it.

These classes provide you getter methods. Using these classes allows you to freely switch between different data formats and don't change your code.

Resolvers solve the problem of data-format migration.

• fields place can be changed
• fields names can be changed

Resolvers can work only with one event/message. It means, if your message has sub-messages it won't work, because resolver will not know with which sub-message should it work.

Implementation advice:

1. raise NotImplementedError -- if your Implementation doesn't support this getter.

Performance impact:

• It a bit slower than using naked field access dict['key'].

2.4. Links

• Report Data Provider
• Th2 Data Services Utils

3. Best practices

Depending on how you work with Data object, it can be slow of fast.
As with a relational database, you can write a query that will return data slowly or quickly, the same when working with a Data object.

Follow the rules to make your work with Data object fast:

1. Use Data.use_cache() if you iterate data more than one time.
2. Try to don't iterate one Data object inside the other one.
   If you should to do it, use short Data object first and long Data object inside the loop.
   It'll allow you open the cache file or create a request to Data source less number of times.

4. Official DataSource implementations

• Lightweight Data Provider Data Source

5. API

If you are looking for classes description see the API Documentation.

6. Examples

• get_started_example.py