dlt.pipeline.helpers

retry_load

def retry_load(retry_on_pipeline_steps: Sequence[TPipelineStep] = (
    "load", )) -> Callable[[BaseException], bool]

View source on GitHub

A retry strategy for Tenacity that, with default setting, will repeat load step for all exceptions that are not terminal

Use this condition with tenacity retry_if_exception. Terminal exceptions are exceptions that will not go away when operations is repeated. Examples: missing configuration values, Authentication Errors, terminally failed jobs exceptions etc.

data = source(...)
for attempt in Retrying(stop=stop_after_attempt(3), retry=retry_if_exception(retry_load(())), reraise=True):
    with attempt:
        p.run(data)

Arguments:

• retry_on_pipeline_steps Tuple[TPipelineStep, ...], optional - which pipeline steps are allowed to be repeated. Default: "load"

retry_schema_update

def retry_schema_update(retry_on_pipeline_steps: Sequence[TPipelineStep] = (
    "load", )) -> Callable[[BaseException], bool]

View source on GitHub

A retry strategy for Tenacity that repeats a pipeline step when a schema update fails.

Use it to retry schema update errors in load step. These typically happen when several pipelines run in parallel and race to create the same table or add the same column. Retrying is safe: dlt re-reads the destination and applies only the changes that are still missing. Compose with retry_load to also retry other load failures.

from tenacity import Retrying, stop_after_attempt, retry_if_exception
from dlt.pipeline.helpers import retry_load, retry_schema_update
should_retry = retry_if_exception(retry_schema_update()) | retry_if_exception(retry_load(("load", "extract")))
for attempt in Retrying(stop=stop_after_attempt(5), retry=should_retry, reraise=True):
    with attempt:
        pipeline.run(data)

Arguments:

• retry_on_pipeline_steps Tuple[TPipelineStep, ...], optional - which pipeline steps are allowed to be repeated. Default: "load"

pipeline_drop Objects

class pipeline_drop()

View source on GitHub

__init__

def __init__(pipeline: "Pipeline",
             resources: Union[Iterable[Union[str, TSimpleRegex]],
                              Union[str, TSimpleRegex]] = (),
             schema_name: Optional[str] = None,
             state_paths: TAnyJsonPath = (),
             drop_all: bool = False,
             state_only: bool = False) -> None

View source on GitHub

Prepares pipeline drop for a particular source/schema and executes it when called. You can inspect the modifications collected during the dry run by inspecting the info property. Drop operation is a special pipeline run with attached commands to drop tables corresponding to resources, associated pipeline state and to replace schema versions in the destination.

Arguments:

• pipeline - Pipeline to drop tables and state from
• resources - List of resources to drop. If empty, no resources are dropped unless drop_all is True
• schema_name - Name of the schema to drop tables from. If not specified, the default schema is used
• state_paths - JSON path(s) relative to the source state to drop
• drop_all - Drop all resources and tables in the schema (supersedes resources list)
• state_only - Drop only state, not tables

prepare_refresh_source

def prepare_refresh_source(
        pipeline: "Pipeline", source: DltSource,
        refresh: TRefreshMode) -> TLoadPackageDropTablesState

View source on GitHub

Prepare refresh mode on the given source, updating the provided schema and pipeline state to reflect tables being dropped/truncated and associated resources state.

Behavior depends in refresh mode:

• drop_sources - all tables generated by resources in source will be dropped, source state will be wiped.
• drop_resources - tables generated by resources that are actually selected in source will be dropped and associated resources stare will be wiped.
• drop_data - tables generated by resources that are actually selected in source will be truncated and associated resources stare will be wiped.

Returns:

The new load package state containing tables that need to be dropped/truncated.