A Hacky Guide to Hive (part 2.2.2: custom_YOson)

in #dev3 months ago (edited)

Context

Please read last post for context.

SQLAlchemy

This is no Python guide, this is no SQL guide.
I just want to get the job done, so I use SQLAlchemy.
You could build the following better or faster, in any language you want, without SQLAlchemy, for a different storage than SQLite... the main thing is: You'll probably need other data.
This is just a way to get to a working demo quickly.

db.py

from typing import List
import datetime
from sqlalchemy import String, ForeignKey, DateTime, create_engine
from sqlalchemy.sql import func
from sqlalchemy.orm import DeclarativeBase, Mapped, mapped_column, relationship, Session

class Base(DeclarativeBase):
    pass

class Block(Base):
    __tablename__ = "blocks"
    num: Mapped[int] = mapped_column(primary_key=True)
    timestamp: Mapped[str] = mapped_column(String(19))    
    sql_timestamp: Mapped[datetime.datetime] = mapped_column(DateTime(timezone=True), server_default=func.now())   
    yos: Mapped[List["Yo"]] = relationship(back_populates="block", cascade="all, delete-orphan")     

class Yo(Base):
    __tablename__ = "yos"
    id: Mapped[int] = mapped_column(primary_key=True, autoincrement=True)
    author: Mapped[str] = mapped_column(String(16))
    tx_id: Mapped[str] = mapped_column(String(40))
    block_num: Mapped[int] = mapped_column(ForeignKey("blocks.num"))
    block: Mapped[Block] = relationship(back_populates="yos")

This is no SQLAlchemy quide.
It is for the most part just the example from the SQLalchemy tutorial, adapted to my data.
It allows me to map SQL tables blocks and yos as classes Block and Yo.
I am not even sure, if I am doing it right, but for the demo it doesn't matter.

def make_db(db):
    Base.metadata.create_all(create_engine(db))

def set_blocks(blocks, session: Session):
    session.add_all(blocks)

def get_last_block(session: Session):    
    return session.query(Block.num).order_by(Block.num.desc()).limit(1).first()[0]    

The main goal of db.py is to be able to:

  • make a DB
  • add blocks to DB table (set)
  • return blocks from DB (get)

For now, I only need to get to the last block, so I built a specialized function, to do just that.

def make_yo_blocks(start, blocks):
    num = start
    yo_blocks = []       
    for block in blocks:                
        yos = []
        for tx_i, transaction in enumerate(block['transactions']):
            for operation in transaction['operations']:
                if operation['type'] == 'custom_json_operation':
                    if operation['value']['id'] == 'YO':
                        yo = Yo(author= operation['value']['required_posting_auths'][0], tx_id= block['transaction_ids'][tx_i])
                        yos.append(yo)
        yo_block = Block(num= num, timestamp= block['timestamp'], yos= yos)
        yo_blocks.append(yo_block)
        num += 1
    return yo_blocks

The above is the last piece.
It does a lot of things for a single function.
It takes the blocks data from the next block and converts it to Blocks (and Yos).
It can break in a lot of places, but that's exactly what it should do for now.

Example

db_file = "sqlite:///yo.db"
init_block = Block(num = 89040473, timestamp = '2024-09-18T12:55:36', yos = [Yo(author = 'felixxx', tx_id = 'eb025cf797ee5bc81d7399282268079cc29cc66d')])

def init():
    make_db(db_file)
    with Session(create_engine(db_file, echo=True)) as session:
        set_blocks([init_block], session)
        session.commit()

If you run this example init twice, you will get an error:
Demonstrating that SQL is already doing stuff; it prevents inserting a block of the same num twice.

custom_YOson.png

custom_YOson

In the final process the api and db handling meet.

api.py

import requests

def get_block_range(start, count, url):
    data = '{"jsonrpc":"2.0", "method":"block_api.get_block_range","params":{"starting_block_num":'+str(start)+',"count": '+str(count)+'},"id":1}'
    return requests.post(url, data)

That's the Hive api from last post. It's really that short.
I modified it just a little, to move data conversion further upstream; Instead of a list of blocks, it now returns raw data.

custom_YOson.py

import db, api

url = 'https://api.hive.blog'
db_file = "sqlite:///yo.db"
count = 10
innit_block = db.Block(num = 89040473, timestamp = '2024-09-18T12:55:36', yos = [db.Yo(author = 'felixxx', tx_id = 'eb025cf797ee5bc81d7399282268079cc29cc66d')])

def innit():
    db.make_db(db_file)
    with db.Session(db.create_engine(db_file)) as session:
        db.set_blocks([innit_block], session)
        session.commit()
    
def tick():    
    with db.Session(db.create_engine(db_file)) as session:        
        last = db.get_last_block(session)
    response = api.get_block_range(last + 1, count, url)
    blocks = response.json()['result']['blocks']
    with db.Session(db.create_engine(db_file)) as session:
        if last == db.get_last_block(session): # cheap way to make this thread-safe        
            db.set_blocks(db.make_yo_blocks(last + 1, blocks), session)
            session.commit()

The main procedure, that should happen during every tick:

  • with open db session:
    • get last block's num from db
  • query Hive node for next (count = 10) blocks
  • convert Hive node response to dictionary, select key 'blocks'
  • with open db session:
    • get last block's num again, if same:
      • convert blocks to Blocks
      • store Blocks

Process

custom_YOson is only 75 lines long:

  • api.py (5)
  • db.py (48)
  • custom_YOson.py (22)

You'd still have to wrap this in a loop, but the core procedure works for stream and resync.
My main goal here was to use functions in tick() that each represent one step of the procedure.

Performance

1 block @ 26 KB10 blocks @ 168 KB100 blocks @ 1919 KB
api.hive.blog: 0.693 s
YOson: 0.020 s
total: 0.714 s
api.hive.blog: 0.976 s
YOson: 0.017 s
total: 0.993 s
api.hive.blog: 7.910 s
YOson: 0.028 s
total: 7.939 s

I did 3 test shots and what's clear: The node performance is the bottleneck.
I could optimize my code or try a different language and database, but I could only shave off miliseconds.
Acessing a node via localhost would probably be substantially faster, but even then: to improve this there probably isn't much you can do in Python...

Conclusion

Hive's state changes. The only way to observe state changes is querying a node.
In many ways, Hive is a like a black box.
Goal was to build a white box, that has healthy data. (yo.db)
How to best sanitize, store and handle the data depends on the application.
You could build something like HiveSQL, hive-engine, a voting service or a game with a yo.db as engine, like above.

Next post I'll cover error handling and deployment and build an application.

Had I understood 8 years ago, when I first found this chain, how to build stuff like above... I could have made a lot of money. A lot of it isn't even Hive specific...

Please test & comment!

Sort:  

Congratulations for your level of mastery. Hats off !

Thanks!
Certainly no master. I have advanced a level above writing long scripts.
Still haven't deployed anything cool.

Have a look at what we are trying to do with OffChain Luxembourg (@offchain-lux) and let me know if you are interested in collaborating. The coolest things come from collaboration IMO

Have a look at what we are trying to do with OffChain Luxembourg

Will do. Thanks!

The coolest things come from collaboration

Not only that.
Alone, I just can't operate any service 24/7.
It takes a few people - whatever the code may be.

thats interesting and i loke that you are caring about the performance from start. great job.

i was wondering how i could use custom_json without having a database. not sure yet what path i will take, but this solution was something like i had in mind. good to see that i was not wrong having a database to query it easier when need past data.