feed range query fix - initial commit

Author: dibahlfi

sdk/cosmos/azure-cosmos/azure/cosmos/_cosmos_client_connection.py

@@ -29,7 +29,7 @@
 import urllib.parse
 import uuid
 from concurrent.futures.thread import ThreadPoolExecutor
-from typing import Callable, Any, Iterable, Mapping, Optional, Sequence, Tuple, Union, cast
+from typing import Callable, Any, Dict, Iterable, Mapping, Optional, Sequence, Tuple, Union, cast
 from typing_extensions import TypedDict
 from urllib3.util.retry import Retry
 
@@ -73,7 +73,18 @@
 from ._read_items_helper import ReadItemsHelperSync
 from ._request_object import RequestObject
 from ._retry_utility import ConnectionRetryPolicy
-from ._routing import routing_map_provider, routing_range
+from ._routing import routing_map_provider
+from ._routing.feed_range_continuation import (
+    _apply_feedrange_request_headers,
+    _build_scope_from_overlaps,
+    _decode_token,
+    _derive_initial_feedranges,
+    _explode_feedrange_on_multi_overlap,
+    _extract_resume_state,
+    _normalize_max_item_count,
+    _set_outbound_continuation,
+    _validate_token_identity,
+)
 from ._query_advisor import get_query_advice_info
 from ._inference_service import _InferenceService
 from .documents import ConnectionPolicy, DatabaseAccount
@@ -3240,7 +3251,17 @@ def __QueryFeed(  # pylint: disable=too-many-locals, too-many-statements, too-ma
         if timeout is not None:
             kwargs.setdefault("timeout", timeout)
 
-        internal_headers_capture = kwargs.pop("_internal_response_headers_capture", None)
+        internal_headers_capture: Optional[Dict[str, Any]] = kwargs.pop(
+            "_internal_response_headers_capture", None
+        )
+
+        def _capture_internal_headers(headers: Mapping[str, Any]) -> None:
+            # Local helper so flow analysis can narrow Optional[Dict] once
+            # and every call site stays a single line.
+            if internal_headers_capture is None:
+                return
+            internal_headers_capture.clear()
+            internal_headers_capture.update(headers)
 
         if query:
             __GetBodiesFromQueryResult = result_fn
@@ -3293,8 +3314,7 @@ def __GetBodiesFromQueryResult(result: dict[str, Any]) -> list[dict[str, Any]]:
             result, last_response_headers = self.__Get(path, request_params, headers, **kwargs)
             self.last_response_headers = last_response_headers
             if internal_headers_capture is not None:
-                internal_headers_capture.clear()
-                internal_headers_capture.update(last_response_headers)
+                _capture_internal_headers(last_response_headers)
             if response_headers_list is not None:
                 response_headers_list.append(last_response_headers.copy())
             if response_hook:
@@ -3348,64 +3368,187 @@ def __GetBodiesFromQueryResult(result: dict[str, Any]) -> list[dict[str, Any]]:
 
         # If feed_range_epk exist, query with the range
         if feed_range_epk is not None:
-            over_lapping_ranges = self._routing_map_provider.get_overlapping_ranges(resource_id, [feed_range_epk],
-                                                                                    options)
-            # It is possible to get more than one over lapping range. We need to get the query results for each one
+            # (a) Look at the continuation the caller passed in.
+            #     - Empty or from a pre-fix SDK: start fresh.
+            #     - One of our v=1 envelopes: check the collection, query, and
+            #       feed_range still match before resuming from it.
+            inbound = _decode_token(options.get("continuation"))
+            if inbound is not None:
+                _validate_token_identity(inbound, resource_id, query, feed_range_epk)
+                current_feedrange, remaining_feedranges, next_backend_cont = _extract_resume_state(inbound)
+            else:
+                # First call (or legacy passthrough). Ask the routing map which
+                # partitions the input feed_range overlaps right now and turn
+                # each overlap into a feedrange (intersection of that partition
+                # and the input feed_range).
+                first_overlaps = self._routing_map_provider.get_overlapping_ranges(
+                    resource_id, [feed_range_epk], options
+                )
+                all_feedranges = _derive_initial_feedranges(feed_range_epk, first_overlaps)
+                if not all_feedranges:
+                    # The input feed_range does not overlap any current
+                    # physical partition. Mirror the async path: stamp
+                    # an empty headers object onto self.last_response_headers
+                    # so callers reading it after a no-op call see a
+                    # consistent (empty) value instead of whatever the
+                    # previous request left behind.
+                    empty_headers: CaseInsensitiveDict = CaseInsensitiveDict()
+                    self.last_response_headers = empty_headers
+                    return [], empty_headers
+                current_feedrange, remaining_feedranges = all_feedranges[0], all_feedranges[1:]
+                next_backend_cont = None
+
+            # (b) max_item_count is the cap for the page we hand back to the
+            # caller. We may need several backend POSTs to fill it (one per
+            # feedrange we have to query). Once the cap is reached we stop;
+            # any feedranges we have not started yet go into the outbound
+            # token so the next call picks up from there.
+            #
+            # Non-positive or non-numeric caps are normalized to "unbounded"
+            # (None) by _normalize_max_item_count: a cap of 0 / -1 would
+            # otherwise make the loop short-circuit before issuing any POST
+            # and emit a continuation token whose current_feedrange is
+            # unchanged - the caller would then pull empty pages forever
+            # without making progress.
+            remaining_budget = _normalize_max_item_count(options.get("maxItemCount"))
+
             results: dict[str, Any] = {}
-            # For each over lapping range we will take a sub range of the feed range EPK that overlaps with the over
-            # lapping physical partition. The EPK sub range will be one of four:
-            # 1) Will have a range min equal to the feed range EPK min, and a range max equal to the over lapping
-            # partition
-            # 2) Will have a range min equal to the over lapping partition range min, and a range max equal to the
-            # feed range EPK range max.
-            # 3) will match exactly with the current over lapping physical partition, so we just return the over lapping
-            # physical partition's partition key id.
-            # 4) Will equal the feed range EPK since it is a sub range of a single physical partition
-            for over_lapping_range in over_lapping_ranges:
-                single_range = routing_range.Range.PartitionKeyRangeToRange(over_lapping_range)
-                # Since the range min and max are all Upper Cased string Hex Values,
-                # we can compare the values lexicographically
-                EPK_sub_range = routing_range.Range(range_min=max(single_range.min, feed_range_epk.min),
-                                                    range_max=min(single_range.max, feed_range_epk.max),
-                                                    isMinInclusive=True, isMaxInclusive=False)
-
-                # set the session token for this specific partition to avoid sending compound token for all partitions
-                base.set_session_token_header(self, req_headers, path, request_params, options,
-                                              over_lapping_range["id"])
-                if single_range.min == EPK_sub_range.min and EPK_sub_range.max == single_range.max:
-                    # The Epk Sub Range spans exactly one physical partition
-                    # In this case we can route to the physical pk range id
-                    req_headers[http_constants.HttpHeaders.PartitionKeyRangeID] = over_lapping_range["id"]
+            last_response_headers: CaseInsensitiveDict = CaseInsensitiveDict()
+
+            while True:
+                if remaining_budget is not None and remaining_budget <= 0:
+                    break  # cap reached; carry the unfinished feedranges forward
+
+                # Look up the live routing map for the current feedrange.
+                # Doing this every iteration is what makes the token
+                # split-safe.
+                overlapping = self._routing_map_provider.get_overlapping_ranges(
+                    resource_id, [current_feedrange], options
+                )
+                overlapping, partition_scope = _build_scope_from_overlaps(
+                    overlapping, current_feedrange
+                )
+
+                # Handle the case where Cosmos split a partition between the
+                # previous run and this one. Example: the saved
+                # current_feedrange used to live inside one partition X, but X
+                # has since been split into children X1 and X2. The routing
+                # map now returns two partitions for the same feedrange. If
+                # we sent one POST to X1 with X's full range as the EPK
+                # filter, the backend would filter in-partition only and
+                # silently drop every row living on X2 (that is how a resume
+                # after a split came back 19 ids short of ground truth in
+                # test_post_split_resume_async).
+                #
+                # So when the lookup returns more than one partition, slice
+                # the saved feedrange into one sub-feedrange per child
+                # (intersection with the saved feedrange, ordered by EPK
+                # min), make the first sub-feedrange the new current one,
+                # put the rest in front of the remaining list, and clear the
+                # saved backend continuation - it was issued by the old
+                # parent partition and the children won't accept it. The next
+                # loop iteration sees a single overlap and falls through to
+                # the normal single-partition POST below.
+                #
+                # Note: if the caller had already pulled some rows from X
+                # before the split, those rows show up again on this resume.
+                # The customer dedupes by document id.
+                current_feedrange, remaining_feedranges, did_explode = _explode_feedrange_on_multi_overlap(
+                    current_feedrange,
+                    overlapping,
+                    remaining_feedranges,
+                )
+                if did_explode:
+                    next_backend_cont = None
+                    overlapping = self._routing_map_provider.get_overlapping_ranges(
+                        resource_id, [current_feedrange], options
+                    )
+                    overlapping, partition_scope = _build_scope_from_overlaps(
+                        overlapping, current_feedrange
+                    )
+
+                sub_options = dict(options)
+                if remaining_budget is not None:
+                    sub_options["maxItemCount"] = remaining_budget
+                if next_backend_cont is not None:
+                    sub_options["continuation"] = next_backend_cont
                 else:
-                    # The Epk Sub Range spans less than a single physical partition
-                    # In this case we route to the physical partition and
-                    # pass the epk sub range to the headers to filter within partition
-                    req_headers[http_constants.HttpHeaders.PartitionKeyRangeID] = over_lapping_range["id"]
-                    req_headers[http_constants.HttpHeaders.StartEpkString] = EPK_sub_range.min
-                    req_headers[http_constants.HttpHeaders.EndEpkString] = EPK_sub_range.max
-                req_headers[http_constants.HttpHeaders.ReadFeedKeyType] = "EffectivePartitionKeyRange"
-                partial_result, last_response_headers = self.__Post(
+                    sub_options.pop("continuation", None)
+
+                # Populate request headers for this single backend POST.
+                # The shared helper handles partition routing (PKR id +
+                # optional EPK filter), page-size cap, and continuation
+                # set/clear so the same rules apply to sync and async.
+                _apply_feedrange_request_headers(
+                    req_headers,
+                    overlapping,
+                    partition_scope,
+                    current_feedrange,
+                    remaining_budget,
+                    sub_options.get("continuation"),
+                )
+                # Use the session token for this specific partition so we don't
+                # send a compound token covering all partitions.
+                base.set_session_token_header(
+                    self, req_headers, path, request_params, sub_options, overlapping[0]["id"]
+                )
+
+                partial_result, sub_response_headers = self.__Post(
                     path, request_params, query, req_headers, **kwargs
                 )
+                last_response_headers = sub_response_headers
                 self.last_response_headers = last_response_headers
                 if internal_headers_capture is not None:
-                    internal_headers_capture.clear()
-                    internal_headers_capture.update(last_response_headers)
-                self._UpdateSessionIfRequired(req_headers, partial_result, last_response_headers)
-                # Introducing a temporary complex function into a critical path to handle aggregated queries
-                # during splits, as a precaution falling back to the original logic if anything goes wrong
+                    _capture_internal_headers(sub_response_headers)
+                self._UpdateSessionIfRequired(req_headers, partial_result, sub_response_headers)
+
+                # Merge results, falling back to a plain extend if the
+                # aggregating merge raises (it can on aggregated queries
+                # during splits).
                 try:
                     results = base._merge_query_results(results, partial_result, query)
-                except Exception: # pylint: disable=broad-exception-caught
-                    # If the new merge logic fails, fall back to the original logic.
+                except Exception:  # pylint: disable=broad-exception-caught
                     if results:
                         results["Documents"].extend(partial_result["Documents"])
                     else:
                         results = partial_result
+
+                items_returned = len(partial_result.get("Documents", []))
+                if remaining_budget is not None:
+                    remaining_budget -= items_returned
                 if response_headers_list is not None:
-                    response_headers_list.append(last_response_headers.copy())
+                    response_headers_list.append(sub_response_headers.copy())
                 if response_hook:
-                    response_hook(last_response_headers, partial_result)
+                    response_hook(sub_response_headers, partial_result)
+
+                next_backend_cont = sub_response_headers.get(http_constants.HttpHeaders.Continuation)
+                if next_backend_cont:
+                    # Current feedrange has more to give. Stay on it; the
+                    # budget check at the top of the next iteration decides
+                    # whether to issue another POST.
+                    continue
+
+                # Current feedrange is drained. Move to the next one if there
+                # is one; otherwise we are done.
+                if not remaining_feedranges:
+                    current_feedrange = None
+                    break
+                current_feedrange = remaining_feedranges.pop(0)
+                next_backend_cont = None
+
+            # (c) Build the outbound token. Clear the continuation header if
+            # there is no work left at all.
+            _set_outbound_continuation(
+                last_response_headers,
+                resource_id,
+                query,
+                feed_range_epk,
+                current_feedrange,
+                remaining_feedranges,
+                next_backend_cont,
+            )
+            self.last_response_headers = last_response_headers
+
             # if the prefix partition query has results lets return it
             if results:
                 if last_response_headers.get(http_constants.HttpHeaders.IndexUtilization) is not None:
@@ -3417,12 +3560,12 @@ def __GetBodiesFromQueryResult(result: dict[str, Any]) -> list[dict[str, Any]]:
                     last_response_headers[http_constants.HttpHeaders.QueryAdvice] = (
                         get_query_advice_info(query_advice_raw))
                 return __GetBodiesFromQueryResult(results), last_response_headers
+            return [], last_response_headers
 
         result, last_response_headers = self.__Post(path, request_params, query, req_headers, **kwargs)
         self.last_response_headers = last_response_headers
         if internal_headers_capture is not None:
-            internal_headers_capture.clear()
-            internal_headers_capture.update(last_response_headers)
+            _capture_internal_headers(last_response_headers)
         self._UpdateSessionIfRequired(req_headers, result, last_response_headers)
         if last_response_headers.get(http_constants.HttpHeaders.IndexUtilization) is not None:
             INDEX_METRICS_HEADER = http_constants.HttpHeaders.IndexUtilization