Add xfail to flaky tests (#1087)

#### Reference Issues/PRs
issue #496

#### What does this implement or fix?
As agreed today, I am adding xfail to the tests that need to be
investigated as part of issue #496 .
The xfails will be removed when the tests are no longer flaky.

python/tests/hypothesis/arcticdb/test_append.py

@@ -135,6 +135,7 @@ def test_(initial: InputFactories, append: InputFactories, match, swap, lmdb_ver
  df_in_str=st.booleans(),
 )
 @SLOW_TESTS_MARK
+@pytest.mark.xfail(reason="Needs to be fixed with issue #496")
 def test_append_with_defragmentation(
  sym,
  col_per_append_df,

python/tests/unit/arcticdb/version_store/test_aggregation.py

@@ -107,6 +107,9 @@ def test_hypothesis_sum_agg(lmdb_version_store, df):
  received_df.replace(-np.inf, np.nan, inplace=True)
  received_df.replace(np.inf, np.nan, inplace=True)
 
+ expected_df.replace(-np.inf, np.nan, inplace=True)
+ expected_df.replace(np.inf, np.nan, inplace=True)
+
  assert_frame_equal(expected_df, received_df)
 
 

python/tests/unit/arcticdb/version_store/test_api.py

@@ -11,6 +11,7 @@
 import pytest
 
 from arcticdb.exceptions import NoSuchVersionException, NoDataFoundException
+from arcticdb.util.test import distinct_timestamps
 
 
 def test_read_descriptor(lmdb_version_store, one_col_df):
@@ -53,25 +54,25 @@ def test_column_names_by_timestamp(lmdb_version_store, one_col_df, two_col_df):
  symbol = "test_column_names_by_timestamp"
 
  # Write a DF with a single column
- lmdb_version_store.write(symbol, one_col_df())
- after_one_col_write = Timestamp.now(tz="UTC")
+ with distinct_timestamps(lmdb_version_store) as first_write_timestamp:
+  lmdb_version_store.write(symbol, one_col_df())
 
  # Ensure the timestamps differ
  time.sleep(0.1)
 
- lmdb_version_store.write(symbol, two_col_df())
- after_two_col_write = Timestamp.now(tz="UTC")
+ with distinct_timestamps(lmdb_version_store) as second_write_timestamp:
+  lmdb_version_store.write(symbol, two_col_df())
 
  # Assert querying with a time before the first write raises an exception
  with pytest.raises(NoDataFoundException) as excinfo:
  lmdb_version_store.column_names(symbol, as_of=Timestamp("1970-01-01", tz="UTC"))
  assert issubclass(excinfo.type, NoSuchVersionException)
 
  # Assert query with the timestamp after the one col write returns only a single column
- assert lmdb_version_store.column_names(symbol, as_of=after_one_col_write) == ["x"]
+ assert lmdb_version_store.column_names(symbol, as_of=first_write_timestamp.after) == ["x"]
 
  # Assert query with the timestamp after the two col write returns two columns
- assert lmdb_version_store.column_names(symbol, as_of=after_two_col_write) == ["x", "y"]
+ assert lmdb_version_store.column_names(symbol, as_of=second_write_timestamp.after) == ["x", "y"]
 
 
 def test_get_num_rows(lmdb_version_store, two_col_df):

python/tests/unit/arcticdb/version_store/test_append.py

@@ -411,6 +411,7 @@ def test_append_mix_ascending_descending(lmdb_version_store):
  assert info["sorted"] == "UNSORTED"
 
 
+@pytest.mark.xfail(reason="Needs to be fixed with issue #496")
 def test_append_with_cont_mem_problem(sym, lmdb_version_store_tiny_segment_dynamic):
  set_config_int("SymbolDataCompact.SegmentCount", 1)
  df0 = pd.DataFrame({"0": ["01234567890123456"]}, index=[pd.Timestamp(0)])

python/tests/unit/arcticdb/version_store/test_filtering.py

@@ -967,7 +967,7 @@ def test_filter_numeric_isin_unsigned(lmdb_version_store):
  df=dataframes_with_names_and_dtypes(["a"], integral_type_strategies()),
  vals=st.frozensets(unsigned_integral_type_strategies(), min_size=1),
 )
-@pytest.mark.skipif(PANDAS_VERSION < Version("1.2"), reason="Early Pandas filtering does not handle unsigned well")
+@pytest.mark.skipif(PANDAS_VERSION < Version("2.0.0"), reason="Early Pandas filtering does not handle unsigned well")
 def test_filter_numeric_isnotin_unsigned(lmdb_version_store, df, vals):
  assume(not df.empty)
  q = QueryBuilder()