fix(concurrency): support failed on http cache write

airbyte_cdk/sources/declarative/concurrent_declarative_source.py

@@ -56,8 +56,9 @@
 
 
 class ConcurrentDeclarativeSource(ManifestDeclarativeSource, Generic[TState]):
-    # By default, we defer to a value of 1 which represents running a connector using the Concurrent CDK engine on only one thread.
-    SINGLE_THREADED_CONCURRENCY_LEVEL = 1
+    # By default, we defer to a value of 2. A value lower than than could cause a PartitionEnqueuer to be stuck in a state of deadlock
+    # because it has hit the limit of futures but not partition reader is consuming them.
+    SINGLE_THREADED_CONCURRENCY_LEVEL = 2
 
     def __init__(
         self,
@@ -121,7 +122,7 @@ def __init__(
             )  # Partition_generation iterates using range based on this value. If this is floored to zero we end up in a dead lock during start up
         else:
             concurrency_level = self.SINGLE_THREADED_CONCURRENCY_LEVEL
-            initial_number_of_partitions_to_generate = self.SINGLE_THREADED_CONCURRENCY_LEVEL
+            initial_number_of_partitions_to_generate = self.SINGLE_THREADED_CONCURRENCY_LEVEL // 2
 
         self._concurrent_source = ConcurrentSource.create(
             num_workers=concurrency_level,

airbyte_cdk/sources/declarative/interpolation/jinja.py

@@ -4,7 +4,7 @@
 
 import ast
 from functools import cache
-from typing import Any, Mapping, Optional, Tuple, Type
+from typing import Any, Mapping, Optional, Set, Tuple, Type
 
 from jinja2 import meta
 from jinja2.environment import Template
@@ -30,6 +30,34 @@ def is_safe_attribute(self, obj: Any, attr: str, value: Any) -> bool:
         return super().is_safe_attribute(obj, attr, value)  # type: ignore  # for some reason, mypy says 'Returning Any from function declared to return "bool"'
 
 
+# These aliases are used to deprecate existing keywords without breaking all existing connectors.
+_ALIASES = {
+    "stream_interval": "stream_slice",  # Use stream_interval to access incremental_sync values
+    "stream_partition": "stream_slice",  # Use stream_partition to access partition router's values
+}
+
+# These extensions are not installed so they're not currently a problem,
+# but we're still explicitely removing them from the jinja context.
+# At worst, this is documentation that we do NOT want to include these extensions because of the potential security risks
+_RESTRICTED_EXTENSIONS = ["jinja2.ext.loopcontrols"]  # Adds support for break continue in loops
+
+# By default, these Python builtin functions are available in the Jinja context.
+# We explicitely remove them because of the potential security risk.
+# Please add a unit test to test_jinja.py when adding a restriction.
+_RESTRICTED_BUILTIN_FUNCTIONS = [
+    "range"
+]  # The range function can cause very expensive computations
+
+_ENVIRONMENT = StreamPartitionAccessEnvironment()
+_ENVIRONMENT.filters.update(**filters)
+_ENVIRONMENT.globals.update(**macros)
+
+for extension in _RESTRICTED_EXTENSIONS:
+    _ENVIRONMENT.extensions.pop(extension, None)
+for builtin in _RESTRICTED_BUILTIN_FUNCTIONS:
+    _ENVIRONMENT.globals.pop(builtin, None)
+
+
 class JinjaInterpolation(Interpolation):
     """
     Interpolation strategy using the Jinja2 template engine.
@@ -48,34 +76,6 @@ class JinjaInterpolation(Interpolation):
     Additional information on jinja templating can be found at https://jinja.palletsprojects.com/en/3.1.x/templates/#
     """
 
-    # These aliases are used to deprecate existing keywords without breaking all existing connectors.
-    ALIASES = {
-        "stream_interval": "stream_slice",  # Use stream_interval to access incremental_sync values
-        "stream_partition": "stream_slice",  # Use stream_partition to access partition router's values
-    }
-
-    # These extensions are not installed so they're not currently a problem,
-    # but we're still explicitely removing them from the jinja context.
-    # At worst, this is documentation that we do NOT want to include these extensions because of the potential security risks
-    RESTRICTED_EXTENSIONS = ["jinja2.ext.loopcontrols"]  # Adds support for break continue in loops
-
-    # By default, these Python builtin functions are available in the Jinja context.
-    # We explicitely remove them because of the potential security risk.
-    # Please add a unit test to test_jinja.py when adding a restriction.
-    RESTRICTED_BUILTIN_FUNCTIONS = [
-        "range"
-    ]  # The range function can cause very expensive computations
-
-    def __init__(self) -> None:
-        self._environment = StreamPartitionAccessEnvironment()
-        self._environment.filters.update(**filters)
-        self._environment.globals.update(**macros)
-
-        for extension in self.RESTRICTED_EXTENSIONS:
-            self._environment.extensions.pop(extension, None)
-        for builtin in self.RESTRICTED_BUILTIN_FUNCTIONS:
-            self._environment.globals.pop(builtin, None)
-
     def eval(
         self,
         input_str: str,
@@ -86,7 +86,7 @@ def eval(
     ) -> Any:
         context = {"config": config, **additional_parameters}
 
-        for alias, equivalent in self.ALIASES.items():
+        for alias, equivalent in _ALIASES.items():
             if alias in context:
                 # This is unexpected. We could ignore or log a warning, but failing loudly should result in fewer surprises
                 raise ValueError(
@@ -105,6 +105,7 @@ def eval(
                 raise Exception(f"Expected a string, got {input_str}")
         except UndefinedError:
             pass
+
         # If result is empty or resulted in an undefined error, evaluate and return the default string
         return self._literal_eval(self._eval(default, context), valid_types)
 
@@ -132,16 +133,16 @@ def _eval(self, s: Optional[str], context: Mapping[str, Any]) -> Optional[str]:
             return s
 
     @cache
-    def _find_undeclared_variables(self, s: Optional[str]) -> Template:
+    def _find_undeclared_variables(self, s: Optional[str]) -> Set[str]:
         """
         Find undeclared variables and cache them
         """
-        ast = self._environment.parse(s)  # type: ignore # parse is able to handle None
+        ast = _ENVIRONMENT.parse(s)  # type: ignore # parse is able to handle None
         return meta.find_undeclared_variables(ast)
 
     @cache
-    def _compile(self, s: Optional[str]) -> Template:
+    def _compile(self, s: str) -> Template:
         """
         We must cache the Jinja Template ourselves because we're using `from_string` instead of a template loader
         """
-        return self._environment.from_string(s)
+        return _ENVIRONMENT.from_string(s)

airbyte_cdk/sources/streams/http/http_client.py

@@ -54,6 +54,7 @@
 from airbyte_cdk.utils.traced_exception import AirbyteTracedException
 
 BODY_REQUEST_METHODS = ("GET", "POST", "PUT", "PATCH")
+logger = logging.getLogger("airbyte")
 
 
 class MessageRepresentationAirbyteTracedErrors(AirbyteTracedException):
@@ -94,6 +95,7 @@ def __init__(
     ):
         self._name = name
         self._api_budget: APIBudget = api_budget or APIBudget(policies=[])
+        self._logger = logger
         if session:
             self._session = session
         else:
@@ -107,7 +109,6 @@ def __init__(
             )
         if isinstance(authenticator, AuthBase):
             self._session.auth = authenticator
-        self._logger = logger
         self._error_handler = error_handler or HttpStatusErrorHandler(self._logger)
         if backoff_strategy is not None:
             if isinstance(backoff_strategy, list):
@@ -139,11 +140,14 @@ def _request_session(self) -> requests.Session:
             # Use in-memory cache if cache_dir is not set
             # This is a non-obvious interface, but it ensures we don't write sql files when running unit tests
             if cache_dir:
+                self._logger.info(f"Using path {cache_dir} for HTTP cache")  # TODO: remove
                 sqlite_path = str(Path(cache_dir) / self.cache_filename)
             else:
+                self._logger.info("Using memory for cache")  # TODO: remove
                 sqlite_path = "file::memory:?cache=shared"
+            backend = SkipFailureSQLiteCache(self._name, sqlite_path)  # TODO maybe add a busy timeout
             return CachedLimiterSession(
-                sqlite_path, backend="sqlite", api_budget=self._api_budget, match_headers=True
+                sqlite_path, backend=backend, api_budget=self._api_budget, match_headers=True
             )  # type: ignore # there are no typeshed stubs for requests_cache
         else:
             return LimiterSession(api_budget=self._api_budget)
@@ -517,3 +521,44 @@ def send_request(
         )
 
         return request, response
+
+
+class SkipFailureSQLiteDict(requests_cache.backends.sqlite.SQLiteDict):
+    def __getitem__(self, key):  # type: ignore  # lib is not typed
+        try:
+            return super().__getitem__(key)  # type: ignore  # lib is not typed
+        except Exception as exception:
+            if not isinstance(exception, KeyError):
+                logger.warning(f"Error while retrieving item from cache: {exception}")
+            else:
+                raise exception
+
+    def _write(self, key: str, value: str) -> None:
+        try:
+            super()._write(key, value)  # type: ignore  # lib is not typed
+        except Exception as exception:
+            logger.warning(f"Error while saving item to cache: {exception}")
+
+
+class SkipFailureSQLiteCache(requests_cache.backends.sqlite.SQLiteCache):
+    def __init__(  # type: ignore  # ignoring as lib is not typed
+        self,
+        table_name="response",
+        db_path="http_cache",
+        serializer=None,
+        **kwargs,
+    ) -> None:
+        super().__init__(db_path, serializer, **kwargs)
+        skwargs = {"serializer": serializer, **kwargs} if serializer else kwargs
+        self.responses: requests_cache.backends.sqlite.SQLiteDict = SkipFailureSQLiteDict(
+            db_path, table_name=table_name, fast_save=True, wal=True, **skwargs
+        )
+        self.redirects: requests_cache.backends.sqlite.SQLiteDict = SkipFailureSQLiteDict(
+            db_path,
+            table_name=f"redirects_{table_name}",
+            fast_save=True,
+            wal=True,
+            lock=self.responses._lock,
+            serializer=None,
+            **kwargs,
+        )

unit_tests/sources/streams/http/test_http_client.py

@@ -1,11 +1,13 @@
 # Copyright (c) 2024 Airbyte, Inc., all rights reserved.
-
+import contextlib
 import logging
 from datetime import timedelta
+from sqlite3 import OperationalError
 from unittest.mock import MagicMock, patch
 
 import pytest
 import requests
+import requests_cache
 from pympler import asizeof
 from requests_cache import CachedRequest
 
@@ -741,3 +743,27 @@ def test_given_different_headers_then_response_is_not_cached(requests_mock):
     )
 
     assert second_response.json()["test"] == "second response"
+
+
+class RaiseOnInsertConnection:
+    def execute(*args, **kwargs) -> None:
+        if "INSERT" in str(args):
+            raise OperationalError("database table is locked")
+
+
+def test_given_cache_save_failure_then_do_not_break(requests_mock, monkeypatch):
+    @contextlib.contextmanager
+    def _create_sqlite_write_error_connection(*args, **kwargs):
+        yield RaiseOnInsertConnection()
+
+    monkeypatch.setattr(
+        requests_cache.backends.sqlite.SQLiteDict,
+        "connection",
+        _create_sqlite_write_error_connection,
+    )
+    http_client = HttpClient(name="test", logger=MagicMock(), use_cache=True)
+    requests_mock.register_uri("GET", "https://google.com/", json={"test": "response"})
+
+    request, response = http_client.send_request("GET", "https://google.com/", request_kwargs={})
+
+    assert response.json()