from pathlib import Path
from typing import (
AbstractSet,
Any,
Callable,
Dict,
Iterator,
Mapping,
Optional,
Sequence,
Union,
cast,
)
import dateutil
from dagster import (
AssetKey,
AssetMaterialization,
AssetObservation,
MetadataValue,
Output,
_check as check,
)
from dagster._core.definitions.metadata import RawMetadataValue
from .types import DbtOutput
# dbt resource types that may be considered assets
ASSET_RESOURCE_TYPES = ["model", "seed", "snapshot"]
def default_node_info_to_asset_key(node_info: Mapping[str, Any]) -> AssetKey:
return AssetKey(node_info["unique_id"].split("."))
def _resource_type(unique_id: str) -> str:
# returns the type of the node (e.g. model, test, snapshot)
return unique_id.split(".")[0]
def _get_input_name(node_info: Mapping[str, Any]) -> str:
# * can be present when sources are sharded tables
return node_info["unique_id"].replace(".", "_").replace("*", "_star")
def _get_output_name(node_info: Mapping[str, Any]) -> str:
return node_info["unique_id"].split(".")[-1]
def _node_result_to_metadata(node_result: Mapping[str, Any]) -> Mapping[str, RawMetadataValue]:
return {
"Materialization Strategy": node_result["config"]["materialized"],
"Database": node_result["database"],
"Schema": node_result["schema"],
"Alias": node_result["alias"],
"Description": node_result["description"],
}
def _timing_to_metadata(timings: Sequence[Mapping[str, Any]]) -> Mapping[str, RawMetadataValue]:
metadata: Dict[str, RawMetadataValue] = {}
for timing in timings:
if timing["name"] == "execute":
desc = "Execution"
elif timing["name"] == "compile":
desc = "Compilation"
else:
continue
# dateutil does not properly expose its modules to static checkers
started_at = dateutil.parser.isoparse(timing["started_at"]) # type: ignore
completed_at = dateutil.parser.isoparse(timing["completed_at"]) # type: ignore
duration = completed_at - started_at
metadata.update(
{
f"{desc} Started At": started_at.isoformat(timespec="seconds"),
f"{desc} Completed At": started_at.isoformat(timespec="seconds"),
f"{desc} Duration": duration.total_seconds(),
}
)
return metadata
def result_to_events(
result: Mapping[str, Any],
docs_url: Optional[str] = None,
node_info_to_asset_key: Optional[Callable[[Mapping[str, Any]], AssetKey]] = None,
manifest_json: Optional[Mapping[str, Any]] = None,
extra_metadata: Optional[Mapping[str, RawMetadataValue]] = None,
generate_asset_outputs: bool = False,
) -> Iterator[Union[AssetMaterialization, AssetObservation, Output]]:
"""This is a hacky solution that attempts to consolidate parsing many of the potential formats
that dbt can provide its results in. This is known to work for CLI Outputs for dbt versions 0.18+,
as well as RPC responses for a similar time period, but as the RPC response schema is not documented
nor enforced, this can become out of date easily.
"""
node_info_to_asset_key = check.opt_callable_param(
node_info_to_asset_key, "node_info_to_asset_key", default=default_node_info_to_asset_key
)
# status comes from set of fields rather than "status"
if "fail" in result:
status = (
"fail"
if result.get("fail")
else "skip"
if result.get("skip")
else "error"
if result.get("error")
else "success"
)
else:
status = result["status"]
# all versions represent timing the same way
metadata = {"Status": status, "Execution Time (seconds)": result["execution_time"]}
metadata.update(_timing_to_metadata(result["timing"]))
# working with a response that contains the node block (RPC and CLI 0.18.x)
if "node" in result:
unique_id = result["node"]["unique_id"]
metadata.update(_node_result_to_metadata(result["node"]))
else:
unique_id = result["unique_id"]
if docs_url:
metadata["docs_url"] = MetadataValue.url(f"{docs_url}#!/model/{unique_id}")
if extra_metadata:
metadata.update(extra_metadata)
# if you have a manifest available, get the full node info, otherwise just populate unique_id
node_info = manifest_json["nodes"][unique_id] if manifest_json else {"unique_id": unique_id}
node_resource_type = _resource_type(unique_id)
if node_resource_type in ASSET_RESOURCE_TYPES and status == "success":
if generate_asset_outputs:
yield Output(
value=None,
output_name=_get_output_name(node_info),
metadata=metadata,
)
else:
yield AssetMaterialization(
asset_key=node_info_to_asset_key(node_info),
description=f"dbt node: {unique_id}",
metadata=metadata,
)
# can only associate tests with assets if we have manifest_json available
elif node_resource_type == "test" and manifest_json and status != "skipped":
upstream_unique_ids = manifest_json["nodes"][unique_id]["depends_on"]["nodes"]
# tests can apply to multiple asset keys
for upstream_id in upstream_unique_ids:
# the upstream id can reference a node or a source
node_info = manifest_json["nodes"].get(upstream_id) or manifest_json["sources"].get(
upstream_id
)
if node_info is None:
continue
upstream_asset_key = node_info_to_asset_key(node_info)
yield AssetObservation(
asset_key=upstream_asset_key,
metadata={
"Test ID": result["unique_id"],
"Test Status": status,
"Test Message": result.get("message") or "",
},
)
def generate_events(
dbt_output: DbtOutput,
node_info_to_asset_key: Optional[Callable[[Mapping[str, Any]], AssetKey]] = None,
manifest_json: Optional[Mapping[str, Any]] = None,
) -> Iterator[Union[AssetMaterialization, AssetObservation]]:
"""This function yields :py:class:`dagster.AssetMaterialization` events for each model updated by
a dbt command, and :py:class:`dagster.AssetObservation` events for each test run.
Information parsed from a :py:class:`~DbtOutput` object.
"""
for result in dbt_output.result["results"]:
for event in result_to_events(
result,
docs_url=dbt_output.docs_url,
node_info_to_asset_key=node_info_to_asset_key,
manifest_json=manifest_json,
):
yield check.inst(
cast(Union[AssetMaterialization, AssetObservation], event),
(AssetMaterialization, AssetObservation),
)
[docs]def generate_materializations(
dbt_output: DbtOutput,
asset_key_prefix: Optional[Sequence[str]] = None,
) -> Iterator[AssetMaterialization]:
"""This function yields :py:class:`dagster.AssetMaterialization` events for each model updated by
a dbt command.
Information parsed from a :py:class:`~DbtOutput` object.
Note that this will not work with output from the `dbt_rpc_resource`, because this resource does
not wait for a response from the RPC server before returning. Instead, use the
`dbt_rpc_sync_resource`, which will wait for execution to complete.
Examples:
.. code-block:: python
from dagster import op, Output
from dagster_dbt.utils import generate_materializations
from dagster_dbt import dbt_cli_resource, dbt_rpc_sync_resource
@op(required_resource_keys={"dbt"})
def my_custom_dbt_run(context):
dbt_output = context.resources.dbt.run()
for materialization in generate_materializations(dbt_output):
# you can modify the materialization object to add extra metadata, if desired
yield materialization
yield Output(my_dbt_output)
@job(resource_defs={{"dbt":dbt_cli_resource}})
def my_dbt_cli_job():
my_custom_dbt_run()
@job(resource_defs={{"dbt":dbt_rpc_sync_resource}})
def my_dbt_rpc_job():
my_custom_dbt_run()
"""
asset_key_prefix = check.opt_sequence_param(asset_key_prefix, "asset_key_prefix", of_type=str)
for event in generate_events(
dbt_output,
node_info_to_asset_key=lambda info: AssetKey(
asset_key_prefix + info["unique_id"].split(".")
),
):
yield check.inst(cast(AssetMaterialization, event), AssetMaterialization)
def select_unique_ids_from_manifest(
select: str,
exclude: str,
state_path: Optional[str] = None,
manifest_json_path: Optional[str] = None,
manifest_json: Optional[Mapping[str, Any]] = None,
) -> AbstractSet[str]:
"""Method to apply a selection string to an existing manifest.json file."""
try:
import dbt.flags as flags
import dbt.graph.cli as graph_cli
import dbt.graph.selector as graph_selector
from dbt.contracts.graph.manifest import Manifest, WritableManifest
from dbt.contracts.state import PreviousState
from dbt.graph import SelectionSpec
from dbt.graph.selector_spec import IndirectSelection
from networkx import DiGraph
except ImportError as e:
raise check.CheckError(
"In order to use the `select` argument on load_assets_from_dbt_manifest, you must have"
"`dbt-core >= 1.0.0` and `networkx` installed."
) from e
if state_path is not None:
previous_state = PreviousState(
path=Path(state_path),
current_path=Path("/tmp/null")
if manifest_json_path is None
else Path(manifest_json_path),
)
else:
previous_state = None
if manifest_json_path is not None:
manifest = WritableManifest.read_and_check_versions(manifest_json_path)
child_map = manifest.child_map
elif manifest_json is not None:
class _DictShim(dict):
"""Shim to enable hydrating a dictionary into a dot-accessible object."""
def __getattr__(self, item):
ret = super().get(item)
# allow recursive access e.g. foo.bar.baz
return _DictShim(ret) if isinstance(ret, dict) else ret
manifest = Manifest(
# dbt expects dataclasses that can be accessed with dot notation, not bare dictionaries
nodes={
unique_id: _DictShim(info) for unique_id, info in manifest_json["nodes"].items()
},
sources={
unique_id: _DictShim(info) for unique_id, info in manifest_json["sources"].items()
},
metrics={
unique_id: _DictShim(info) for unique_id, info in manifest_json["metrics"].items()
},
exposures={
unique_id: _DictShim(info) for unique_id, info in manifest_json["exposures"].items()
},
)
child_map = manifest_json["child_map"]
else:
check.failed("Must provide either a manifest_json_path or manifest_json.")
graph = graph_selector.Graph(DiGraph(incoming_graph_data=child_map))
# create a parsed selection from the select string
flags.INDIRECT_SELECTION = IndirectSelection.Eager
parsed_spec: SelectionSpec = graph_cli.parse_union([select], True)
if exclude:
parsed_spec = graph_cli.SelectionDifference(
components=[parsed_spec, graph_cli.parse_union([exclude], True)]
)
# execute this selection against the graph
selector = graph_selector.NodeSelector(graph, manifest, previous_state=previous_state)
selected, _ = selector.select_nodes(parsed_spec)
return selected