Skip to content

graflo.architecture.query

DB-agnostic read contract: what an agent may ask of a live graph.

Layer 2. Imports filter/ and graph_types/ (layer 1) and nothing from db/ — enforcement has to be expressible without a driver, or "enforced in core" would mean "enforced wherever a driver happens to be imported".

Not in contract/, which is the write-side manifest at layer 3: mixing a read contract into it would muddy a boundary the layering test already guards.

AggregateQuery

Bases: GraphQuery

Count or summarise one vertex type.

Source code in graflo/architecture/query/models.py 
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
class AggregateQuery(GraphQuery):
    """Count or summarise one vertex type."""

    vertex_type: str = PydanticField(...)
    function: AggregationType = PydanticField(
        default=AggregationType.COUNT, description="Aggregation to apply."
    )
    group_by: str | None = PydanticField(
        default=None, description="Property to group by. COUNT only."
    )
    aggregated_field: str | None = PydanticField(
        default=None,
        description="Property to aggregate. Required for everything but COUNT.",
    )
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def aggregation(self) -> AggregationType:
        """`function` as an enum. See `NeighborQuery.edge_direction`."""
        return AggregationType(self.function)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.function != AggregationType.COUNT and not self.aggregated_field:
            raise ValueError(
                f"aggregated_field is required for {self.function}; only COUNT "
                "can aggregate without naming a property"
            )
        if self.group_by and self.function != AggregationType.COUNT:
            raise ValueError(
                f"group_by is only supported for COUNT, not {self.function}"
            )

aggregation property

function as an enum. See NeighborQuery.edge_direction.

CapExceededError

Bases: ValueError

A query asked for more than a cap allows.

Carries the cap's name so the surface can say which limit was hit rather than returning a generic validation failure — an agent that is told "too many hops, max is 3" can retry; one told "invalid request" cannot.

Source code in graflo/architecture/query/caps.py 
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
class CapExceededError(ValueError):
    """A query asked for more than a cap allows.

    Carries the cap's name so the surface can say *which* limit was hit rather
    than returning a generic validation failure — an agent that is told "too
    many hops, max is 3" can retry; one told "invalid request" cannot.
    """

    def __init__(self, cap: str, requested: object, allowed: object) -> None:
        self.cap = cap
        self.requested = requested
        self.allowed = allowed
        super().__init__(
            f"{cap} exceeded: requested {requested!r}, maximum is {allowed!r}"
        )

GraphQuery

Bases: ConfigBaseModel

Shared envelope for every read request.

Note

There is intentionally no caps field. Caps come from :data:HARD_CAPS and are lowered only through :meth:narrowed, so a request body cannot raise its own ceiling.

Source code in graflo/architecture/query/models.py 
 31
 32
 33
 34
 35
 36
 37
 38
 39
 40
 41
 42
 43
 44
 45
 46
 47
 48
 49
 50
 51
 52
 53
 54
 55
 56
 57
 58
 59
 60
 61
 62
 63
 64
 65
 66
 67
 68
 69
 70
 71
 72
 73
 74
 75
 76
 77
 78
 79
 80
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
class GraphQuery(ConfigBaseModel):
    """Shared envelope for every read request.

    Note:
        There is intentionally no ``caps`` field. Caps come from
        :data:`HARD_CAPS` and are lowered only through :meth:`narrowed`, so a
        request body cannot raise its own ceiling.
    """

    limit: int = PydanticField(
        default=100, ge=1, description="Rows to return, within `max_rows`."
    )
    timeout_s: float = PydanticField(
        default=10.0, gt=0, description="Seconds to allow, within `timeout_s`."
    )
    projection: list[str] | None = PydanticField(
        default=None,
        description=(
            "Property names to return. None returns whatever the backend "
            "stores, subject to the cap's allow-list."
        ),
    )

    def finish_init(self, caps: QueryCaps | None = None) -> Self:
        """Validate against *caps*, defaulting to the core ceiling.

        Returns:
            Self, so the call composes: ``NodeQuery(...).finish_init()``.

        Raises:
            CapExceededError: Naming the cap that was exceeded.
        """
        effective = caps or HARD_CAPS
        if self.limit > effective.max_rows:
            raise CapExceededError("max_rows", self.limit, effective.max_rows)
        if self.timeout_s > effective.timeout_s:
            raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
        if self.projection is not None and effective.projection_allow_list is not None:
            permitted = set(effective.projection_allow_list)
            denied = sorted(n for n in self.projection if n not in permitted)
            if denied:
                raise CapExceededError(
                    "projection_allow_list", denied, effective.projection_allow_list
                )
        self._validate_specific(effective)
        return self

    def _validate_specific(self, caps: QueryCaps) -> None:
        """Per-query-kind checks. Overridden by subclasses."""

    def narrowed(self, caps: QueryCaps) -> Self:
        """Return a copy fitted to *caps*, raising on anything explicitly asked for.

        The split is what makes caps both enforceable and usable:

        - A value the caller **explicitly set** above a cap raises
          :class:`CapExceededError`. Silently clamping it would hand back a
          partial answer the caller believes is complete — the failure mode this
          codebase already rejected for TigerGraph edge direction.
        - A value the caller **left at its default** is clamped. A policy of
          ``max_rows=5`` must not 422 every request that simply did not mention
          a limit; that would make strict policies unusable rather than safe.

        ``model_fields_set`` is what distinguishes the two, which is precisely
        why it must be consulted rather than comparing against the default.

        Projection is always intersected rather than raising, since an
        allow-list exists to *hide* properties: refusing the request would tell
        the caller which forbidden property they guessed correctly.
        """
        effective = HARD_CAPS.narrow(caps)
        explicit = self.model_fields_set
        narrowed = self.model_copy(deep=True)

        if "limit" in explicit and self.limit > effective.max_rows:
            raise CapExceededError("max_rows", self.limit, effective.max_rows)
        narrowed.limit = min(self.limit, effective.max_rows)

        if "timeout_s" in explicit and self.timeout_s > effective.timeout_s:
            raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
        narrowed.timeout_s = min(self.timeout_s, effective.timeout_s)

        if (
            narrowed.projection is not None
            and effective.projection_allow_list is not None
        ):
            permitted = set(effective.projection_allow_list)
            narrowed.projection = [n for n in narrowed.projection if n in permitted]

        narrowed._narrow_specific(effective, explicit)
        return narrowed

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        """Per-query-kind fitting. Overridden by subclasses.

        Args:
            caps: The effective ceiling.
            explicit: Field names the caller actually set, so an explicit
                over-ask can raise while a default is clamped.
        """

finish_init(caps=None)

Validate against caps, defaulting to the core ceiling.

Returns:

Type Description
Self

Self, so the call composes: NodeQuery(...).finish_init().

Raises:

Type Description
CapExceededError

Naming the cap that was exceeded.
Source code in graflo/architecture/query/models.py 
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
def finish_init(self, caps: QueryCaps | None = None) -> Self:
    """Validate against *caps*, defaulting to the core ceiling.

    Returns:
        Self, so the call composes: ``NodeQuery(...).finish_init()``.

    Raises:
        CapExceededError: Naming the cap that was exceeded.
    """
    effective = caps or HARD_CAPS
    if self.limit > effective.max_rows:
        raise CapExceededError("max_rows", self.limit, effective.max_rows)
    if self.timeout_s > effective.timeout_s:
        raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
    if self.projection is not None and effective.projection_allow_list is not None:
        permitted = set(effective.projection_allow_list)
        denied = sorted(n for n in self.projection if n not in permitted)
        if denied:
            raise CapExceededError(
                "projection_allow_list", denied, effective.projection_allow_list
            )
    self._validate_specific(effective)
    return self

narrowed(caps)

Return a copy fitted to caps, raising on anything explicitly asked for.

The split is what makes caps both enforceable and usable:

  • A value the caller explicitly set above a cap raises :class:CapExceededError. Silently clamping it would hand back a partial answer the caller believes is complete — the failure mode this codebase already rejected for TigerGraph edge direction.
  • A value the caller left at its default is clamped. A policy of max_rows=5 must not 422 every request that simply did not mention a limit; that would make strict policies unusable rather than safe.

model_fields_set is what distinguishes the two, which is precisely why it must be consulted rather than comparing against the default.

Projection is always intersected rather than raising, since an allow-list exists to hide properties: refusing the request would tell the caller which forbidden property they guessed correctly.

Source code in graflo/architecture/query/models.py 
 81
 82
 83
 84
 85
 86
 87
 88
 89
 90
 91
 92
 93
 94
 95
 96
 97
 98
 99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
def narrowed(self, caps: QueryCaps) -> Self:
    """Return a copy fitted to *caps*, raising on anything explicitly asked for.

    The split is what makes caps both enforceable and usable:

    - A value the caller **explicitly set** above a cap raises
      :class:`CapExceededError`. Silently clamping it would hand back a
      partial answer the caller believes is complete — the failure mode this
      codebase already rejected for TigerGraph edge direction.
    - A value the caller **left at its default** is clamped. A policy of
      ``max_rows=5`` must not 422 every request that simply did not mention
      a limit; that would make strict policies unusable rather than safe.

    ``model_fields_set`` is what distinguishes the two, which is precisely
    why it must be consulted rather than comparing against the default.

    Projection is always intersected rather than raising, since an
    allow-list exists to *hide* properties: refusing the request would tell
    the caller which forbidden property they guessed correctly.
    """
    effective = HARD_CAPS.narrow(caps)
    explicit = self.model_fields_set
    narrowed = self.model_copy(deep=True)

    if "limit" in explicit and self.limit > effective.max_rows:
        raise CapExceededError("max_rows", self.limit, effective.max_rows)
    narrowed.limit = min(self.limit, effective.max_rows)

    if "timeout_s" in explicit and self.timeout_s > effective.timeout_s:
        raise CapExceededError("timeout_s", self.timeout_s, effective.timeout_s)
    narrowed.timeout_s = min(self.timeout_s, effective.timeout_s)

    if (
        narrowed.projection is not None
        and effective.projection_allow_list is not None
    ):
        permitted = set(effective.projection_allow_list)
        narrowed.projection = [n for n in narrowed.projection if n in permitted]

    narrowed._narrow_specific(effective, explicit)
    return narrowed

NeighborQuery

Bases: GraphQuery

What is adjacent to one anchor vertex.

The instance-plane counterpart of SchemaGraph.schema_neighbors. They are different questions and must never share a name: this one asks which rows are adjacent, the other which types can be.

Source code in graflo/architecture/query/models.py 
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
class NeighborQuery(GraphQuery):
    """What is adjacent to one anchor vertex.

    The instance-plane counterpart of ``SchemaGraph.schema_neighbors``. They are
    different questions and must never share a name: this one asks which *rows*
    are adjacent, the other which *types* can be.
    """

    vertex_type: str = PydanticField(..., description="Logical type of the anchor.")
    key: str | dict[str, Any] = PydanticField(
        ..., description="Anchor identity value, or a single-field mapping."
    )
    hops: int = PydanticField(
        default=1, ge=1, description="Hop distance, within `max_hops`."
    )
    direction: EdgeDirection = PydanticField(
        default=EdgeDirection.OUT,
        description=(
            "Orientation followed from the anchor. Defaults to OUT, matching "
            "`Connection.fetch_edges`; an edge declared `directed: false` is "
            "followed both ways regardless."
        ),
    )
    edge_relations: list[str] | None = PydanticField(
        default=None, description="Restrict to these relations. None means all."
    )
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def edge_direction(self) -> EdgeDirection:
        """`direction` as an enum, for handing to a driver.

        `ConfigBaseModel` sets ``use_enum_values=True``, so the stored value is
        the bare string. Backends compare against `EdgeDirection` members, and a
        string silently matches none of them — which each backend then resolves
        differently, so the same query returns different neighbourhoods
        depending on where it runs. Always cross the boundary through this.
        """
        return EdgeDirection(self.direction)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.hops > caps.max_hops:
            raise CapExceededError("max_hops", self.hops, caps.max_hops)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        if "hops" in explicit and self.hops > caps.max_hops:
            raise CapExceededError("max_hops", self.hops, caps.max_hops)
        self.hops = min(self.hops, caps.max_hops)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

edge_direction property

direction as an enum, for handing to a driver.

ConfigBaseModel sets use_enum_values=True, so the stored value is the bare string. Backends compare against EdgeDirection members, and a string silently matches none of them — which each backend then resolves differently, so the same query returns different neighbourhoods depending on where it runs. Always cross the boundary through this.

NodeQuery

Bases: GraphQuery

List vertices of one type, optionally filtered.

Source code in graflo/architecture/query/models.py 
133
134
135
136
137
138
139
class NodeQuery(GraphQuery):
    """List vertices of one type, optionally filtered."""

    vertex_type: str = PydanticField(..., description="Logical vertex type name.")
    filters: FilterExpression | None = PydanticField(
        default=None, description="Predicate applied before limiting."
    )

QueryCaps

Bases: ConfigBaseModel

Ceilings a read request must fit inside.

Defaults are the core ceiling. A deployment narrows them per connection via :meth:GraphQuery.narrowed; nothing widens them.

Source code in graflo/architecture/query/caps.py 
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
class QueryCaps(ConfigBaseModel):
    """Ceilings a read request must fit inside.

    Defaults are the core ceiling. A deployment narrows them per connection via
    :meth:`GraphQuery.narrowed`; nothing widens them.
    """

    max_hops: int = PydanticField(
        default=3, ge=1, description="Deepest traversal permitted."
    )
    max_rows: int = PydanticField(
        default=1000, ge=1, description="Most rows a single query may return."
    )
    max_elements: int = PydanticField(
        default=5000,
        ge=1,
        description="Most vertices plus edges a response may carry.",
    )
    timeout_s: float = PydanticField(
        default=30.0, gt=0, description="Longest a query may run."
    )
    max_edge_types: int = PydanticField(
        default=20,
        ge=1,
        description="Most distinct relations one request may name.",
    )
    max_seeds: int = PydanticField(
        default=10,
        ge=1,
        description="Most anchor vertices a traversal may start from.",
    )
    projection_allow_list: list[str] | None = PydanticField(
        default=None,
        description=(
            "Property names a response may include. None means unrestricted; "
            "an empty list means nothing may be projected, which is not the "
            "same thing."
        ),
    )

    def narrow(self, other: QueryCaps) -> QueryCaps:
        """Combine two cap sets, taking the stricter of each.

        Narrowing is a lattice meet, not an override: a policy that tried to
        raise a ceiling silently becomes a no-op rather than a privilege
        escalation.
        """
        allow: list[str] | None
        if self.projection_allow_list is None:
            allow = other.projection_allow_list
        elif other.projection_allow_list is None:
            allow = self.projection_allow_list
        else:
            # Intersection, order fixed for reproducibility.
            permitted = set(other.projection_allow_list)
            allow = [n for n in self.projection_allow_list if n in permitted]
        return QueryCaps(
            max_hops=min(self.max_hops, other.max_hops),
            max_rows=min(self.max_rows, other.max_rows),
            max_elements=min(self.max_elements, other.max_elements),
            timeout_s=min(self.timeout_s, other.timeout_s),
            max_edge_types=min(self.max_edge_types, other.max_edge_types),
            max_seeds=min(self.max_seeds, other.max_seeds),
            projection_allow_list=allow,
        )

narrow(other)

Combine two cap sets, taking the stricter of each.

Narrowing is a lattice meet, not an override: a policy that tried to raise a ceiling silently becomes a no-op rather than a privilege escalation.

Source code in graflo/architecture/query/caps.py 
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
def narrow(self, other: QueryCaps) -> QueryCaps:
    """Combine two cap sets, taking the stricter of each.

    Narrowing is a lattice meet, not an override: a policy that tried to
    raise a ceiling silently becomes a no-op rather than a privilege
    escalation.
    """
    allow: list[str] | None
    if self.projection_allow_list is None:
        allow = other.projection_allow_list
    elif other.projection_allow_list is None:
        allow = self.projection_allow_list
    else:
        # Intersection, order fixed for reproducibility.
        permitted = set(other.projection_allow_list)
        allow = [n for n in self.projection_allow_list if n in permitted]
    return QueryCaps(
        max_hops=min(self.max_hops, other.max_hops),
        max_rows=min(self.max_rows, other.max_rows),
        max_elements=min(self.max_elements, other.max_elements),
        timeout_s=min(self.timeout_s, other.timeout_s),
        max_edge_types=min(self.max_edge_types, other.max_edge_types),
        max_seeds=min(self.max_seeds, other.max_seeds),
        projection_allow_list=allow,
    )

QueryResult

Bases: ConfigBaseModel

A query's answer plus the honest caveats.

Source code in graflo/architecture/query/result.py 
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
class QueryResult(ConfigBaseModel):
    """A query's answer plus the honest caveats."""

    container: GraphContainer = PydanticField(
        default_factory=GraphContainer,
        description="DB-agnostic vertices and edges, identical across backends.",
    )
    element_count: int = PydanticField(
        default=0, ge=0, description="Vertices plus edges carried."
    )
    truncated: bool = PydanticField(
        default=False, description="Whether a cap cut the answer short."
    )
    caps_hit: list[str] = PydanticField(
        default_factory=list,
        description=(
            "Which caps bound this answer, by name. Empty when nothing bound "
            "it — the only way a caller can tell a complete answer from one "
            "that happens to fit."
        ),
    )
    elapsed_ms: int = PydanticField(default=0, ge=0)

    @classmethod
    def of(
        cls,
        container: GraphContainer,
        *,
        caps_hit: list[str] | None = None,
        elapsed_ms: int = 0,
    ) -> QueryResult:
        """Build a result, deriving the element count from the container."""
        hit = caps_hit or []
        count = sum(len(docs) for docs in container.vertices.values()) + sum(
            len(rows) for rows in container.edges.values()
        )
        return cls(
            container=container,
            element_count=count,
            truncated=bool(hit),
            caps_hit=hit,
            elapsed_ms=elapsed_ms,
        )

of(container, *, caps_hit=None, elapsed_ms=0) classmethod

Build a result, deriving the element count from the container.

Source code in graflo/architecture/query/result.py 
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
@classmethod
def of(
    cls,
    container: GraphContainer,
    *,
    caps_hit: list[str] | None = None,
    elapsed_ms: int = 0,
) -> QueryResult:
    """Build a result, deriving the element count from the container."""
    hit = caps_hit or []
    count = sum(len(docs) for docs in container.vertices.values()) + sum(
        len(rows) for rows in container.edges.values()
    )
    return cls(
        container=container,
        element_count=count,
        truncated=bool(hit),
        caps_hit=hit,
        elapsed_ms=elapsed_ms,
    )

TraverseQuery

Bases: GraphQuery

What is reachable from a set of anchors.

Source code in graflo/architecture/query/models.py 
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
class TraverseQuery(GraphQuery):
    """What is reachable from a set of anchors."""

    seeds: list[dict[str, Any]] = PydanticField(
        ...,
        min_length=1,
        description=(
            "Anchors, each `{vertex_type, key}`. Bounded by `max_seeds`: a "
            "traversal fans out per seed, so seed count multiplies cost."
        ),
    )
    max_hops: int = PydanticField(default=2, ge=1)
    direction: EdgeDirection = PydanticField(default=EdgeDirection.ANY)
    edge_relations: list[str] | None = PydanticField(default=None)
    filters: FilterExpression | None = PydanticField(default=None)

    @property
    def edge_direction(self) -> EdgeDirection:
        """`direction` as an enum. See `NeighborQuery.edge_direction`."""
        return EdgeDirection(self.direction)

    def _validate_specific(self, caps: QueryCaps) -> None:
        if self.max_hops > caps.max_hops:
            raise CapExceededError("max_hops", self.max_hops, caps.max_hops)
        if len(self.seeds) > caps.max_seeds:
            raise CapExceededError("max_seeds", len(self.seeds), caps.max_seeds)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )
        for seed in self.seeds:
            missing = {"vertex_type", "key"} - set(seed)
            if missing:
                raise ValueError(
                    f"seed {seed!r} is missing {sorted(missing)}; each seed needs "
                    "a vertex_type and a key"
                )

    def _narrow_specific(self, caps: QueryCaps, explicit: set[str]) -> None:
        if "max_hops" in explicit and self.max_hops > caps.max_hops:
            raise CapExceededError("max_hops", self.max_hops, caps.max_hops)
        self.max_hops = min(self.max_hops, caps.max_hops)
        # Seeds are always explicit — there is no default set of anchors — so
        # dropping any of them would silently answer a different question.
        if len(self.seeds) > caps.max_seeds:
            raise CapExceededError("max_seeds", len(self.seeds), caps.max_seeds)
        if self.edge_relations and len(self.edge_relations) > caps.max_edge_types:
            raise CapExceededError(
                "max_edge_types", len(self.edge_relations), caps.max_edge_types
            )

edge_direction property

direction as an enum. See NeighborQuery.edge_direction.