UncountablePythonSDK 0.0.24__py3-none-any.whl → 0.0.131__py3-none-any.whl

uncountable/integration/queue_runner/datastore/datastore_sqlite.py

@@ -0,0 +1,250 @@
+import datetime
+import uuid
+from datetime import UTC
+
+from sqlalchemy import delete, insert, or_, select, text, update
+from sqlalchemy.engine import Engine
+
+from pkgs.argument_parser import CachedParser
+from pkgs.serialization_util import serialize_for_storage
+from uncountable.integration.db.session import DBSessionMaker
+from uncountable.integration.queue_runner.datastore.interface import Datastore
+from uncountable.integration.queue_runner.datastore.model import Base, QueuedJob
+from uncountable.types import queued_job_t
+
+queued_job_payload_parser = CachedParser(queued_job_t.QueuedJobPayload)
+
+MAX_QUEUE_WINDOW_DAYS = 30
+
+
+class DatastoreSqlite(Datastore):
+    def __init__(self, session_maker: DBSessionMaker) -> None:
+        self.session_maker = session_maker
+        super().__init__()
+
+    @classmethod
+    def setup(cls, engine: Engine) -> None:
+        Base.metadata.create_all(engine)
+        with engine.connect() as connection:
+            if not bool(
+                connection.execute(
+                    text(
+                        "select exists (select 1 from pragma_table_info('queued_jobs') where name='status');"
+                    )
+                ).scalar()
+            ):
+                connection.execute(
+                    text("alter table queued_jobs add column status VARCHAR")
+                )
+
+    def add_job_to_queue(
+        self, job_payload: queued_job_t.QueuedJobPayload, job_ref_name: str
+    ) -> queued_job_t.QueuedJob:
+        with self.session_maker() as session:
+            serialized_payload = serialize_for_storage(job_payload)
+            queued_job_uuid = str(uuid.uuid4())
+            num_attempts = 0
+            submitted_at = datetime.datetime.now(UTC)
+            insert_stmt = insert(QueuedJob).values({
+                QueuedJob.id.key: queued_job_uuid,
+                QueuedJob.job_ref_name.key: job_ref_name,
+                QueuedJob.payload.key: serialized_payload,
+                QueuedJob.status.key: queued_job_t.JobStatus.QUEUED,
+                QueuedJob.num_attempts: num_attempts,
+                QueuedJob.submitted_at: submitted_at,
+            })
+            session.execute(insert_stmt)
+            return queued_job_t.QueuedJob(
+                queued_job_uuid=queued_job_uuid,
+                job_ref_name=job_ref_name,
+                payload=job_payload,
+                status=queued_job_t.JobStatus.QUEUED,
+                submitted_at=submitted_at,
+                num_attempts=num_attempts,
+            )
+
+    def retry_job(
+        self,
+        queued_job_uuid: str,
+    ) -> queued_job_t.QueuedJob | None:
+        with self.session_maker() as session:
+            select_stmt = select(
+                QueuedJob.id,
+                QueuedJob.payload,
+                QueuedJob.num_attempts,
+                QueuedJob.job_ref_name,
+                QueuedJob.status,
+                QueuedJob.submitted_at,
+            ).filter(QueuedJob.id == queued_job_uuid)
+            existing_job = session.execute(select_stmt).one_or_none()
+
+            if (
+                existing_job is None
+                or existing_job.status != queued_job_t.JobStatus.FAILED
+            ):
+                return None
+
+            update_stmt = (
+                update(QueuedJob)
+                .values({QueuedJob.status.key: queued_job_t.JobStatus.QUEUED})
+                .filter(QueuedJob.id == queued_job_uuid)
+            )
+            session.execute(update_stmt)
+
+            return queued_job_t.QueuedJob(
+                queued_job_uuid=existing_job.id,
+                job_ref_name=existing_job.job_ref_name,
+                num_attempts=existing_job.num_attempts,
+                status=queued_job_t.JobStatus.QUEUED,
+                submitted_at=existing_job.submitted_at,
+                payload=queued_job_payload_parser.parse_storage(existing_job.payload),
+            )
+
+    def increment_num_attempts(self, queued_job_uuid: str) -> int:
+        with self.session_maker() as session:
+            update_stmt = (
+                update(QueuedJob)
+                .values({QueuedJob.num_attempts.key: QueuedJob.num_attempts + 1})
+                .filter(QueuedJob.id == queued_job_uuid)
+            )
+            session.execute(update_stmt)
+            session.flush()
+            # IMPROVE: python3's sqlite does not support the RETURNING clause
+            select_stmt = select(QueuedJob.num_attempts).filter(
+                QueuedJob.id == queued_job_uuid
+            )
+            return int(session.execute(select_stmt).one().num_attempts)
+
+    def remove_job_from_queue(self, queued_job_uuid: str) -> None:
+        with self.session_maker() as session:
+            delete_stmt = delete(QueuedJob).filter(QueuedJob.id == queued_job_uuid)
+            session.execute(delete_stmt)
+
+    def update_job_status(
+        self, queued_job_uuid: str, status: queued_job_t.JobStatus
+    ) -> None:
+        with self.session_maker() as session:
+            update_stmt = (
+                update(QueuedJob)
+                .values({QueuedJob.status.key: status})
+                .filter(QueuedJob.id == queued_job_uuid)
+            )
+            session.execute(update_stmt)
+
+    def list_queued_job_metadata(
+        self, offset: int = 0, limit: int | None = 100
+    ) -> list[queued_job_t.QueuedJobMetadata]:
+        with self.session_maker() as session:
+            select_statement = (
+                select(
+                    QueuedJob.id,
+                    QueuedJob.job_ref_name,
+                    QueuedJob.num_attempts,
+                    QueuedJob.status,
+                    QueuedJob.submitted_at,
+                )
+                .order_by(QueuedJob.submitted_at)
+                .offset(offset)
+                .limit(limit)
+            )
+
+            queued_job_metadata: list[queued_job_t.QueuedJobMetadata] = [
+                queued_job_t.QueuedJobMetadata(
+                    queued_job_uuid=row.id,
+                    job_ref_name=row.job_ref_name,
+                    num_attempts=row.num_attempts,
+                    status=row.status or queued_job_t.JobStatus.QUEUED,
+                    submitted_at=row.submitted_at,
+                )
+                for row in session.execute(select_statement)
+            ]
+
+            return queued_job_metadata
+
+    def get_next_queued_job_for_ref_name(
+        self, job_ref_name: str
+    ) -> queued_job_t.QueuedJob | None:
+        with self.session_maker() as session:
+            select_stmt = (
+                select(
+                    QueuedJob.id,
+                    QueuedJob.payload,
+                    QueuedJob.num_attempts,
+                    QueuedJob.job_ref_name,
+                    QueuedJob.status,
+                    QueuedJob.submitted_at,
+                )
+                .filter(QueuedJob.job_ref_name == job_ref_name)
+                .filter(
+                    or_(
+                        QueuedJob.status == queued_job_t.JobStatus.QUEUED,
+                        QueuedJob.status.is_(None),
+                    )
+                )
+                .limit(1)
+                .order_by(QueuedJob.submitted_at)
+            )
+
+            for row in session.execute(select_stmt):
+                parsed_payload = queued_job_payload_parser.parse_storage(row.payload)
+                return queued_job_t.QueuedJob(
+                    queued_job_uuid=row.id,
+                    job_ref_name=row.job_ref_name,
+                    num_attempts=row.num_attempts,
+                    status=row.status or queued_job_t.JobStatus.QUEUED,
+                    submitted_at=row.submitted_at,
+                    payload=parsed_payload,
+                )
+
+            return None
+
+    def load_job_queue(self) -> list[queued_job_t.QueuedJob]:
+        with self.session_maker() as session:
+            select_stmt = (
+                select(
+                    QueuedJob.id,
+                    QueuedJob.payload,
+                    QueuedJob.num_attempts,
+                    QueuedJob.job_ref_name,
+                    QueuedJob.status,
+                    QueuedJob.submitted_at,
+                )
+                .filter(
+                    or_(
+                        QueuedJob.status == queued_job_t.JobStatus.QUEUED,
+                        QueuedJob.status.is_(None),
+                    )
+                )
+                .order_by(QueuedJob.submitted_at)
+            )
+
+            queued_jobs: list[queued_job_t.QueuedJob] = []
+            for row in session.execute(select_stmt):
+                parsed_payload = queued_job_payload_parser.parse_storage(row.payload)
+                queued_jobs.append(
+                    queued_job_t.QueuedJob(
+                        queued_job_uuid=row.id,
+                        job_ref_name=row.job_ref_name,
+                        num_attempts=row.num_attempts,
+                        status=row.status or queued_job_t.JobStatus.QUEUED,
+                        submitted_at=row.submitted_at,
+                        payload=parsed_payload,
+                    )
+                )
+
+            return queued_jobs
+
+    def vaccuum_queued_jobs(self) -> None:
+        with self.session_maker() as session:
+            delete_stmt = (
+                delete(QueuedJob)
+                .filter(QueuedJob.status == queued_job_t.JobStatus.QUEUED)
+                .filter(
+                    QueuedJob.submitted_at
+                    <= (
+                        datetime.datetime.now(UTC)
+                        - datetime.timedelta(days=MAX_QUEUE_WINDOW_DAYS)
+                    )
+                )
+            )
+            session.execute(delete_stmt)

uncountable/integration/queue_runner/datastore/interface.py

@@ -0,0 +1,29 @@
+from abc import ABC, abstractmethod
+
+from uncountable.types import queued_job_t
+
+
+class Datastore(ABC):
+    @abstractmethod
+    def add_job_to_queue(
+        self, job_payload: queued_job_t.QueuedJobPayload, job_ref_name: str
+    ) -> queued_job_t.QueuedJob: ...
+
+    @abstractmethod
+    def remove_job_from_queue(self, queued_job_uuid: str) -> None: ...
+
+    @abstractmethod
+    def increment_num_attempts(self, queued_job_uuid: str) -> int: ...
+
+    @abstractmethod
+    def load_job_queue(self) -> list[queued_job_t.QueuedJob]: ...
+
+    @abstractmethod
+    def get_next_queued_job_for_ref_name(
+        self, job_ref_name: str
+    ) -> queued_job_t.QueuedJob | None: ...
+
+    @abstractmethod
+    def list_queued_job_metadata(
+        self, offset: int, limit: int | None
+    ) -> list[queued_job_t.QueuedJobMetadata]: ...

uncountable/integration/queue_runner/datastore/model.py

@@ -0,0 +1,24 @@
+from sqlalchemy import JSON, BigInteger, Column, DateTime, Enum, Text
+from sqlalchemy.orm import declarative_base
+from sqlalchemy.sql import func
+
+from uncountable.types import queued_job_t
+
+Base = declarative_base()
+
+
+class QueuedJob(Base):
+    __tablename__ = "queued_jobs"
+
+    id = Column(Text, primary_key=True)
+    job_ref_name = Column(Text, nullable=False, index=True)
+    submitted_at = Column(
+        DateTime(timezone=True), server_default=func.current_timestamp(), nullable=False
+    )
+    payload = Column(JSON, nullable=False)
+    num_attempts = Column(BigInteger, nullable=False, default=0, server_default="0")
+    status = Column(
+        Enum(queued_job_t.JobStatus, length=None),
+        default=queued_job_t.JobStatus.QUEUED,
+        nullable=True,
+    )

uncountable/integration/queue_runner/job_scheduler.py

@@ -0,0 +1,200 @@
+import asyncio
+import sys
+import typing
+from concurrent.futures import ProcessPoolExecutor
+from dataclasses import dataclass
+
+from opentelemetry.trace import get_current_span
+
+from uncountable.integration.queue_runner.command_server import (
+    CommandEnqueueJob,
+    CommandEnqueueJobResponse,
+    CommandQueue,
+    CommandRetryJob,
+    CommandRetryJobResponse,
+    CommandTask,
+)
+from uncountable.integration.queue_runner.command_server.types import (
+    CommandVaccuumQueuedJobs,
+)
+from uncountable.integration.queue_runner.datastore import DatastoreSqlite
+from uncountable.integration.queue_runner.datastore.interface import Datastore
+from uncountable.integration.queue_runner.worker import Worker
+from uncountable.integration.scan_profiles import load_profiles
+from uncountable.integration.telemetry import Logger
+from uncountable.types import job_definition_t, queued_job_t
+
+from .types import ResultQueue, ResultTask
+
+_MAX_JOB_WORKERS = 5
+
+
+@dataclass(kw_only=True, frozen=True)
+class JobListenerKey:
+    profile_name: str
+    subqueue_name: str = "default"
+
+
+def _get_job_worker_key(
+    job_definition: job_definition_t.JobDefinition, profile_name: str
+) -> JobListenerKey:
+    if job_definition.subqueue_name is not None:
+        return JobListenerKey(
+            profile_name=profile_name, subqueue_name=job_definition.subqueue_name
+        )
+    return JobListenerKey(profile_name=profile_name)
+
+
+def on_worker_crash(
+    worker_key: JobListenerKey,
+) -> typing.Callable[[asyncio.Task], None]:
+    def hook(task: asyncio.Task) -> None:
+        Logger(get_current_span()).log_exception(
+            Exception(
+                f"worker {worker_key.profile_name}_{worker_key.subqueue_name} crashed unexpectedly"
+            )
+        )
+        sys.exit(1)
+
+    return hook
+
+
+def _start_workers(
+    process_pool: ProcessPoolExecutor, result_queue: ResultQueue, datastore: Datastore
+) -> dict[str, Worker]:
+    profiles = load_profiles()
+    job_queue_worker_lookup: dict[JobListenerKey, Worker] = {}
+    job_worker_lookup: dict[str, Worker] = {}
+    job_definition_lookup: dict[str, job_definition_t.JobDefinition] = {}
+    for profile in profiles:
+        for job_definition in profile.jobs:
+            job_definition_lookup[job_definition.id] = job_definition
+            job_worker_key = _get_job_worker_key(job_definition, profile.name)
+            if job_worker_key not in job_queue_worker_lookup:
+                worker = Worker(
+                    process_pool=process_pool,
+                    listen_queue=asyncio.Queue(),
+                    result_queue=result_queue,
+                    datastore=datastore,
+                )
+                task = asyncio.create_task(worker.run_worker_loop())
+                task.add_done_callback(on_worker_crash(job_worker_key))
+                job_queue_worker_lookup[job_worker_key] = worker
+            job_worker_lookup[job_definition.id] = job_queue_worker_lookup[
+                job_worker_key
+            ]
+    return job_worker_lookup
+
+
+async def start_scheduler(
+    command_queue: CommandQueue, datastore: DatastoreSqlite
+) -> None:
+    logger = Logger(get_current_span())
+    result_queue: ResultQueue = asyncio.Queue()
+
+    with ProcessPoolExecutor(max_workers=_MAX_JOB_WORKERS) as process_pool:
+        job_worker_lookup = _start_workers(
+            process_pool, result_queue, datastore=datastore
+        )
+
+        queued_jobs = datastore.load_job_queue()
+
+        async def enqueue_queued_job(queued_job: queued_job_t.QueuedJob) -> None:
+            try:
+                worker = job_worker_lookup[queued_job.job_ref_name]
+            except KeyError as e:
+                logger.log_exception(e)
+                datastore.update_job_status(
+                    queued_job.queued_job_uuid, queued_job_t.JobStatus.FAILED
+                )
+                return
+            await worker.listen_queue.put(queued_job)
+
+        async def _enqueue_or_deduplicate_job(
+            job_ref_name: str,
+            payload: queued_job_t.QueuedJobPayload,
+        ) -> str:
+            if isinstance(
+                payload.invocation_context,
+                (
+                    queued_job_t.InvocationContextCron,
+                    queued_job_t.InvocationContextManual,
+                ),
+            ):
+                existing_queued_job = datastore.get_next_queued_job_for_ref_name(
+                    job_ref_name=job_ref_name
+                )
+                if existing_queued_job is not None:
+                    return existing_queued_job.queued_job_uuid
+            queued_job = datastore.add_job_to_queue(
+                job_payload=payload,
+                job_ref_name=job_ref_name,
+            )
+            await enqueue_queued_job(queued_job)
+            return queued_job.queued_job_uuid
+
+        async def _handle_enqueue_job_command(command: CommandEnqueueJob) -> None:
+            queued_job_uuid = await _enqueue_or_deduplicate_job(
+                job_ref_name=command.job_ref_name,
+                payload=command.payload,
+            )
+            await command.response_queue.put(
+                CommandEnqueueJobResponse(queued_job_uuid=queued_job_uuid)
+            )
+
+        async def _handle_retry_job_command(command: CommandRetryJob) -> None:
+            queued_job = datastore.retry_job(command.queued_job_uuid)
+            if queued_job is None:
+                await command.response_queue.put(
+                    CommandRetryJobResponse(queued_job_uuid=None)
+                )
+                return
+
+            await enqueue_queued_job(queued_job)
+            await command.response_queue.put(
+                CommandRetryJobResponse(queued_job_uuid=queued_job.queued_job_uuid)
+            )
+
+        def _handle_vaccuum_queued_jobs_command(
+            command: CommandVaccuumQueuedJobs,
+        ) -> None:
+            logger.log_info("Vaccuuming queued jobs...")
+            datastore.vaccuum_queued_jobs()
+
+        for queued_job in queued_jobs:
+            await enqueue_queued_job(queued_job)
+
+        result_task: ResultTask = asyncio.create_task(result_queue.get())
+        command_task: CommandTask = asyncio.create_task(command_queue.get())
+        while True:
+            finished, _ = await asyncio.wait(
+                [result_task, command_task], return_when=asyncio.FIRST_COMPLETED
+            )
+
+            for task in finished:
+                if task == command_task:
+                    command = command_task.result()
+                    match command:
+                        case CommandEnqueueJob():
+                            await _handle_enqueue_job_command(command=command)
+                        case CommandRetryJob():
+                            await _handle_retry_job_command(command=command)
+                        case CommandVaccuumQueuedJobs():
+                            _handle_vaccuum_queued_jobs_command(command=command)
+                        case _:
+                            typing.assert_never(command)
+                    command_task = asyncio.create_task(command_queue.get())
+                elif task == result_task:
+                    queued_job_result = result_task.result()
+                    match queued_job_result.job_result.success:
+                        case True:
+                            datastore.update_job_status(
+                                queued_job_result.queued_job_uuid,
+                                queued_job_t.JobStatus.SUCCESS,
+                            )
+                        case False:
+                            datastore.update_job_status(
+                                queued_job_result.queued_job_uuid,
+                                queued_job_t.JobStatus.FAILED,
+                            )
+                    result_task = asyncio.create_task(result_queue.get())

uncountable/integration/queue_runner/queue_runner.py

@@ -0,0 +1,34 @@
+import asyncio
+
+from uncountable.integration.db.connect import IntegrationDBService, create_db_engine
+from uncountable.integration.db.session import get_session_maker
+from uncountable.integration.queue_runner.command_server import serve
+from uncountable.integration.queue_runner.command_server.types import CommandQueue
+from uncountable.integration.queue_runner.datastore import DatastoreSqlite
+from uncountable.integration.queue_runner.job_scheduler import start_scheduler
+
+
+async def queue_runner_loop() -> None:
+    command_queue: CommandQueue = asyncio.Queue()
+    engine = create_db_engine(IntegrationDBService.RUNNER)
+    session_maker = get_session_maker(engine)
+
+    datastore = DatastoreSqlite(session_maker)
+    datastore.setup(engine)
+
+    command_server = asyncio.create_task(serve(command_queue, datastore))
+
+    scheduler = asyncio.create_task(start_scheduler(command_queue, datastore))
+
+    await scheduler
+    await command_server
+
+
+def start_queue_runner() -> None:
+    loop = asyncio.new_event_loop()
+    loop.run_until_complete(queue_runner_loop())
+    loop.close()
+
+
+if __name__ == "__main__":
+    start_queue_runner()

uncountable/integration/queue_runner/types.py

@@ -0,0 +1,7 @@
+from asyncio import Queue, Task
+
+from uncountable.types import queued_job_t
+
+ListenQueue = Queue[queued_job_t.QueuedJob]
+ResultQueue = Queue[queued_job_t.QueuedJobResult]
+ResultTask = Task[queued_job_t.QueuedJobResult]

uncountable/integration/queue_runner/worker.py

@@ -0,0 +1,116 @@
+import asyncio
+from concurrent.futures import ProcessPoolExecutor
+from dataclasses import dataclass
+
+from opentelemetry.trace import get_current_span
+
+from uncountable.core.async_batch import AsyncBatchProcessor
+from uncountable.integration.construct_client import construct_uncountable_client
+from uncountable.integration.executors.executors import execute_job
+from uncountable.integration.job import JobArguments
+from uncountable.integration.queue_runner.datastore.interface import Datastore
+from uncountable.integration.queue_runner.types import ListenQueue, ResultQueue
+from uncountable.integration.scan_profiles import load_profiles
+from uncountable.integration.telemetry import JobLogger, Logger, get_otel_tracer
+from uncountable.types import base_t, job_definition_t, queued_job_t
+
+
+class Worker:
+    def __init__(
+        self,
+        *,
+        process_pool: ProcessPoolExecutor,
+        listen_queue: ListenQueue,
+        result_queue: ResultQueue,
+        datastore: Datastore,
+    ) -> None:
+        self.process_pool = process_pool
+        self.listen_queue = listen_queue
+        self.result_queue = result_queue
+        self.datastore = datastore
+
+    async def run_worker_loop(self) -> None:
+        logger = Logger(get_current_span())
+        while True:
+            try:
+                queued_job = await self.listen_queue.get()
+                self.datastore.increment_num_attempts(queued_job.queued_job_uuid)
+                loop = asyncio.get_event_loop()
+                result = await loop.run_in_executor(
+                    self.process_pool, run_queued_job, queued_job
+                )
+                assert isinstance(result, job_definition_t.JobResult)
+                await self.result_queue.put(
+                    queued_job_t.QueuedJobResult(
+                        job_result=result, queued_job_uuid=queued_job.queued_job_uuid
+                    )
+                )
+            except BaseException as e:
+                logger.log_exception(e)
+                raise e
+
+
+@dataclass(kw_only=True)
+class RegisteredJobDetails:
+    profile_metadata: job_definition_t.ProfileMetadata
+    job_definition: job_definition_t.JobDefinition
+
+
+def get_registered_job_details(job_ref_name: str) -> RegisteredJobDetails:
+    profiles = load_profiles()
+    for profile_metadata in profiles:
+        for job_definition in profile_metadata.jobs:
+            if job_definition.id == job_ref_name:
+                return RegisteredJobDetails(
+                    profile_metadata=profile_metadata,
+                    job_definition=job_definition,
+                )
+    raise Exception(f"profile not found for job {job_ref_name}")
+
+
+def _resolve_queued_job_payload(queued_job: queued_job_t.QueuedJob) -> base_t.JsonValue:
+    match queued_job.payload.invocation_context:
+        case queued_job_t.InvocationContextCron():
+            return None
+        case queued_job_t.InvocationContextManual():
+            return None
+        case queued_job_t.InvocationContextWebhook():
+            return queued_job.payload.invocation_context.webhook_payload
+
+
+def run_queued_job(
+    queued_job: queued_job_t.QueuedJob,
+) -> job_definition_t.JobResult:
+    with get_otel_tracer().start_as_current_span(name="run_queued_job") as span:
+        job_details = get_registered_job_details(queued_job.job_ref_name)
+        job_logger = JobLogger(
+            base_span=span,
+            profile_metadata=job_details.profile_metadata,
+            job_definition=job_details.job_definition,
+        )
+        try:
+            client = construct_uncountable_client(
+                profile_meta=job_details.profile_metadata, logger=job_logger
+            )
+            batch_processor = AsyncBatchProcessor(client=client)
+
+            payload = _resolve_queued_job_payload(queued_job)
+
+            args = JobArguments(
+                job_definition=job_details.job_definition,
+                client=client,
+                batch_processor=batch_processor,
+                profile_metadata=job_details.profile_metadata,
+                logger=job_logger,
+                payload=payload,
+                job_uuid=queued_job.queued_job_uuid,
+            )
+
+            return execute_job(
+                args=args,
+                profile_metadata=job_details.profile_metadata,
+                job_definition=job_details.job_definition,
+            )
+        except BaseException as e:
+            job_logger.log_exception(e)
+            return job_definition_t.JobResult(success=False)