def test_function_failure():
env = get_env()
g = Graph(env)
s = env._local_python_storage
# Initial graph
batches = 2
cfg = {"batches": batches, "fail": True}
source = g.create_node(customer_source, params=cfg)
blocks = produce(source, graph=g, target_storage=s, env=env)
assert len(blocks) == 1
records = blocks[0].as_records()
assert len(records) == 2
with env.md_api.begin():
assert env.md_api.count(select(DataFunctionLog)) == 1
assert env.md_api.count(select(DataBlockLog)) == 1
pl = env.md_api.execute(select(DataFunctionLog)).scalar_one_or_none()
assert pl.node_key == source.key
assert pl.graph_id == g.get_metadata_obj().hash
assert pl.node_start_state == {}
assert pl.node_end_state == {"records_imported": chunk_size}
assert pl.function_key == source.function.key
assert pl.function_params == cfg
assert pl.error is not None
assert FAIL_MSG in pl.error["error"]
ns = env.md_api.execute(
select(NodeState).filter(NodeState.node_key == pl.node_key)
).scalar_one_or_none()
assert ns.state == {"records_imported": chunk_size}
# Run again without failing, should see different result
source.params["fail"] = False
blocks = produce(source, graph=g, target_storage=s, env=env)
assert len(blocks) == 1
records = blocks[0].as_records()
assert len(records) == batch_size
with env.md_api.begin():
assert env.md_api.count(select(DataFunctionLog)) == 2
assert env.md_api.count(select(DataBlockLog)) == 2
pl = (
env.md_api.execute(
select(DataFunctionLog).order_by(DataFunctionLog.completed_at.desc())
)
.scalars()
.first()
)
assert pl.node_key == source.key
assert pl.graph_id == g.get_metadata_obj().hash
assert pl.node_start_state == {"records_imported": chunk_size}
assert pl.node_end_state == {"records_imported": chunk_size + batch_size}
assert pl.function_key == source.function.key
assert pl.function_params == cfg
assert pl.error is None
ns = env.md_api.execute(
select(NodeState).filter(NodeState.node_key == pl.node_key)
).scalar_one_or_none()
assert ns.state == {"records_imported": chunk_size + batch_size}
def test_exe():
env = make_test_env()
g = Graph(env)
node = g.create_node(key="node", snap=snap_t1_source)
exe = env.get_executable(node)
result = ExecutionManager(exe).execute()
with env.md_api.begin():
assert not result.output_blocks
assert env.md_api.count(select(SnapLog)) == 1
pl = env.md_api.execute(select(SnapLog)).scalar_one_or_none()
assert pl.node_key == node.key
assert pl.graph_id == g.get_metadata_obj().hash
assert pl.node_start_state == {}
assert pl.node_end_state == {}
assert pl.snap_key == node.snap.key
assert pl.snap_params == {}
class TestStreams:
def setup(self):
ctx = make_test_run_context()
self.ctx = ctx
self.env = ctx.env
self.g = Graph(self.env)
self.graph = self.g.get_metadata_obj()
self.dr1t1 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema1",
realized_schema_key="_test.TestSchema1",
)
self.dr2t1 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema1",
realized_schema_key="_test.TestSchema1",
)
self.dr1t2 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema2",
realized_schema_key="_test.TestSchema2",
)
self.dr2t2 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema2",
realized_schema_key="_test.TestSchema2",
)
self.node_source = self.g.create_node(key="pipe_source",
pipe=pipe_t1_source)
self.node1 = self.g.create_node(key="pipe1",
pipe=pipe_t1_sink,
upstream="pipe_source")
self.node2 = self.g.create_node(key="pipe2",
pipe=pipe_t1_to_t2,
upstream="pipe_source")
self.node3 = self.g.create_node(key="pipe3",
pipe=pipe_generic,
upstream="pipe_source")
self.sess = self.env._get_new_metadata_session()
self.sess.add(self.dr1t1)
self.sess.add(self.dr2t1)
self.sess.add(self.dr1t2)
self.sess.add(self.dr2t2)
self.sess.add(self.graph)
def teardown(self):
self.sess.close()
def test_stream_unprocessed_pristine(self):
s = StreamBuilder(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query(self.ctx, self.sess).first() is None
def test_stream_unprocessed_eligible(self):
dfl = PipeLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
pipe_key=self.node_source.pipe.key,
runtime_url="test",
)
drl = DataBlockLog(
pipe_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.sess.add_all([dfl, drl])
s = StreamBuilder(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query(self.ctx, self.sess).first() == self.dr1t1
def test_stream_unprocessed_ineligible_already_input(self):
dfl = PipeLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
pipe_key=self.node_source.pipe.key,
runtime_url="test",
)
drl = DataBlockLog(
pipe_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
dfl2 = PipeLog(
graph_id=self.graph.hash,
node_key=self.node1.key,
pipe_key=self.node1.pipe.key,
runtime_url="test",
)
drl2 = DataBlockLog(
pipe_log=dfl2,
data_block=self.dr1t1,
direction=Direction.INPUT,
)
self.sess.add_all([dfl, drl, dfl2, drl2])
s = StreamBuilder(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query(self.ctx, self.sess).first() is None
def test_stream_unprocessed_ineligible_already_output(self):
"""
By default we don't input a block that has already been output by a DF, _even if that block was never input_,
UNLESS input is a self reference (`this`). This is to prevent infinite loops.
"""
dfl = PipeLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
pipe_key=self.node_source.pipe.key,
runtime_url="test",
)
drl = DataBlockLog(
pipe_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
dfl2 = PipeLog(
graph_id=self.graph.hash,
node_key=self.node1.key,
pipe_key=self.node1.pipe.key,
runtime_url="test",
)
drl2 = DataBlockLog(
pipe_log=dfl2,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.sess.add_all([dfl, drl, dfl2, drl2])
s = StreamBuilder(nodes=self.node_source)
s1 = s.filter_unprocessed(self.node1)
assert s1.get_query(self.ctx, self.sess).first() is None
# But ok with self reference
s2 = s.filter_unprocessed(self.node1, allow_cycle=True)
assert s2.get_query(self.ctx, self.sess).first() == self.dr1t1
def test_stream_unprocessed_eligible_schema(self):
dfl = PipeLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
pipe_key=self.node_source.pipe.key,
runtime_url="test",
)
drl = DataBlockLog(
pipe_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.sess.add_all([dfl, drl])
s = StreamBuilder(nodes=self.node_source, schema="TestSchema1")
s = s.filter_unprocessed(self.node1)
assert s.get_query(self.ctx, self.sess).first() == self.dr1t1
s = StreamBuilder(nodes=self.node_source, schema="TestSchema2")
s = s.filter_unprocessed(self.node1)
assert s.get_query(self.ctx, self.sess).first() is None
def test_operators(self):
dfl = PipeLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
pipe_key=self.node_source.pipe.key,
runtime_url="test",
)
drl = DataBlockLog(
pipe_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
drl2 = DataBlockLog(
pipe_log=dfl,
data_block=self.dr2t1,
direction=Direction.OUTPUT,
)
self.sess.add_all([dfl, drl, drl2])
self._cnt = 0
@operator
def count(stream: DataBlockStream) -> DataBlockStream:
for db in stream:
self._cnt += 1
yield db
sb = StreamBuilder(nodes=self.node_source)
expected_cnt = sb.get_query(self.ctx, self.sess).count()
assert expected_cnt == 2
list(count(sb).as_managed_stream(self.ctx, self.sess))
assert self._cnt == expected_cnt
# Test composed operators
self._cnt = 0
list(count(latest(sb)).as_managed_stream(self.ctx, self.sess))
assert self._cnt == 1
# Test kwargs
self._cnt = 0
list(
count(filter(sb, function=lambda db: False)).as_managed_stream(
self.ctx, self.sess))
assert self._cnt == 0
class TestStreams:
def setup(self):
ctx = make_test_run_context()
self.ctx = ctx
self.env = ctx.env
self.sess = self.env.md_api.begin()
self.sess.__enter__()
self.g = Graph(self.env)
self.graph = self.g.get_metadata_obj()
self.dr1t1 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema1",
realized_schema_key="_test.TestSchema1",
)
self.dr2t1 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema1",
realized_schema_key="_test.TestSchema1",
)
self.dr1t2 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema2",
realized_schema_key="_test.TestSchema2",
)
self.dr2t2 = DataBlockMetadata(
nominal_schema_key="_test.TestSchema2",
realized_schema_key="_test.TestSchema2",
)
self.node_source = self.g.create_node(key="snap_source",
snap=snap_t1_source)
self.node1 = self.g.create_node(key="snap1",
snap=snap_t1_sink,
input="snap_source")
self.node2 = self.g.create_node(key="snap2",
snap=snap_t1_to_t2,
input="snap_source")
self.node3 = self.g.create_node(key="snap3",
snap=snap_generic,
input="snap_source")
self.env.md_api.add(self.dr1t1)
self.env.md_api.add(self.dr2t1)
self.env.md_api.add(self.dr1t2)
self.env.md_api.add(self.dr2t2)
self.env.md_api.add(self.graph)
def teardown(self):
self.sess.__exit__(None, None, None)
def test_stream_unprocessed_pristine(self):
s = stream(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query_result(self.env).scalar_one_or_none() is None
def test_stream_unprocessed_eligible(self):
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.env.md_api.add_all([dfl, drl])
s = stream(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query_result(self.env).scalar_one_or_none() == self.dr1t1
def test_stream_unprocessed_ineligible_already_input(self):
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
dfl2 = SnapLog(
graph_id=self.graph.hash,
node_key=self.node1.key,
snap_key=self.node1.snap.key,
runtime_url="test",
)
drl2 = DataBlockLog(
snap_log=dfl2,
data_block=self.dr1t1,
direction=Direction.INPUT,
)
self.env.md_api.add_all([dfl, drl, dfl2, drl2])
s = stream(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
assert s.get_query_result(self.env).scalar_one_or_none() is None
def test_stream_unprocessed_ineligible_already_output(self):
"""
By default we don't input a block that has already been output by a DF, _even if that block was never input_,
UNLESS input is a self reference (`this`). This is to prevent infinite loops.
"""
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
dfl2 = SnapLog(
graph_id=self.graph.hash,
node_key=self.node1.key,
snap_key=self.node1.snap.key,
runtime_url="test",
)
drl2 = DataBlockLog(
snap_log=dfl2,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.env.md_api.add_all([dfl, drl, dfl2, drl2])
s = stream(nodes=self.node_source)
s1 = s.filter_unprocessed(self.node1)
assert s1.get_query_result(self.env).scalar_one_or_none() is None
# But ok with self reference
s2 = s.filter_unprocessed(self.node1, allow_cycle=True)
assert s2.get_query_result(self.env).scalar_one_or_none() == self.dr1t1
def test_stream_unprocessed_eligible_schema(self):
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
self.env.md_api.add_all([dfl, drl])
s = stream(nodes=self.node_source, schema="TestSchema1")
s = s.filter_unprocessed(self.node1)
assert s.get_query_result(self.env).scalar_one_or_none() == self.dr1t1
s = stream(nodes=self.node_source, schema="TestSchema2")
s = s.filter_unprocessed(self.node1)
assert s.get_query_result(self.env).scalar_one_or_none() is None
def test_operators(self):
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
drl2 = DataBlockLog(
snap_log=dfl,
data_block=self.dr2t1,
direction=Direction.OUTPUT,
)
self.env.md_api.add_all([dfl, drl, drl2])
self._cnt = 0
@operator
def count(stream: DataBlockStream) -> DataBlockStream:
for db in stream:
self._cnt += 1
yield db
sb = stream(nodes=self.node_source)
expected_cnt = sb.get_count(self.env)
assert expected_cnt == 2
list(count(sb).as_managed_stream(self.ctx))
assert self._cnt == expected_cnt
# Test composed operators
self._cnt = 0
list(count(latest(sb)).as_managed_stream(self.ctx))
assert self._cnt == 1
# Test kwargs
self._cnt = 0
list(
count(filter(sb, function=lambda db: False)).as_managed_stream(
self.ctx))
assert self._cnt == 0
def test_managed_stream(self):
dfl = SnapLog(
graph_id=self.graph.hash,
node_key=self.node_source.key,
snap_key=self.node_source.snap.key,
runtime_url="test",
)
drl = DataBlockLog(
snap_log=dfl,
data_block=self.dr1t1,
direction=Direction.OUTPUT,
)
dfl2 = SnapLog(
graph_id=self.graph.hash,
node_key=self.node1.key,
snap_key=self.node1.snap.key,
runtime_url="test",
)
drl2 = DataBlockLog(
snap_log=dfl2,
data_block=self.dr1t1,
direction=Direction.INPUT,
)
self.env.md_api.add_all([dfl, drl, dfl2, drl2])
s = stream(nodes=self.node_source)
s = s.filter_unprocessed(self.node1)
ctx = make_test_run_context()
with ctx.env.md_api.begin():
dbs = ManagedDataBlockStream(ctx, stream_builder=s)
with pytest.raises(StopIteration):
assert next(dbs) is None
