1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
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
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
131
132
133
134
135
136
137
138
139
140
141
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
//! Shared execution context — holds all state shared across executor sub-components.
//!
//! [`ExecutorContext`] is the single source of truth for the schema, adapter,
//! configuration, and caches used during query execution. It is always accessed
//! via `Arc<ExecutorContext>` so sub-executors can be cheaply cloned.
use std::{collections::HashMap, sync::Arc};
use moka::sync::Cache as MokaCache;
use super::{QueryType, support::relay::RelayDispatch};
use crate::{
backend::{
AdminSqlOutcome, AdminSqlRequest, ResultCacheStats,
traits::DatabaseAdapter,
types::{DatabaseType, PoolMetrics, QueryStatEntry},
},
cache::ViewName,
error::Result,
graphql::ParsedQuery,
runtime::{QueryMatcher, QueryPlanner, RuntimeConfig},
schema::{CompiledSchema, IntrospectionResponses},
};
/// All shared state for an executor instance.
///
/// Constructed once at `Executor::new()` / `Executor::with_config()` and then
/// stored as `Arc<ExecutorContext>`. Sub-executors (query runner, mutation
/// runner, etc.) each hold a clone of this `Arc`.
pub(super) struct ExecutorContext {
/// Compiled schema with optimized SQL templates.
pub(super) schema: CompiledSchema,
/// The compiled schema's version (its content hash), computed **once** here
/// at construction (`CompiledSchema::content_hash()` re-serialises + hashes
/// the whole schema — far too expensive to call per-mutation). Stamped onto
/// every change-log outbox row's `schema_version` column so a row records the
/// deployment that produced it (the #378 replay / zero-downtime correctness
/// handle). A per-deployment constant — it changes on any schema change.
pub(super) schema_version: Arc<str>,
/// How each nested object field's rows are gated by its type's row security —
/// decided once, here, from the schema and the RLS policy's declared paths.
pub(super) nested_row_gates: super::runners::query_nested::NestedRowGates,
/// Shared database adapter for query execution.
pub(super) adapter: Arc<dyn DatabaseAdapter>,
/// The adapter's type, named by the constructor that still knew it — the erased
/// `adapter` cannot say — and carried over by a rebuild with the adapter itself.
pub(super) adapter_name: &'static str,
/// Type-erased **write** capability slot.
///
/// The write handle: `Some` exactly when the executor was built through a constructor
/// bounded on [`Writer`](crate::backend::Writer) — implementing `execute_write` is the
/// capability (ruling AA 6).
///
/// Every write dispatch takes its handle from here, so **obtaining the handle is
/// the check**. That is the property a boolean does not have: there is no way to
/// reach the database and forget to consult it. It is the shape `relay` next door
/// already uses.
pub(super) writer: Option<Arc<dyn crate::backend::Writer>>,
/// Type-erased relay capability slot.
///
/// `Some` when constructed via `new_with_relay`. `None` returns a
/// `FraiseQLError::Validation` for relay queries — no `unreachable!()`.
pub(super) relay: Option<Arc<dyn RelayDispatch>>,
/// Query matching engine (stateless).
pub(super) matcher: QueryMatcher,
/// Query execution planner (stateless).
pub(super) planner: QueryPlanner,
/// Runtime configuration (timeouts, complexity limits, etc.).
pub(super) config: RuntimeConfig,
/// Pre-built introspection responses cached for `__schema` and `__type` queries.
pub(super) introspection: IntrospectionResponses,
/// O(1) lookup index for Relay `node(id)` queries.
pub(super) node_type_index: HashMap<String, Arc<str>>,
/// GATE-1 query-structure validator (depth / complexity / alias / size),
/// resolved once at construction: the embedder-installed
/// `RuntimeConfig::query_validation` when set, otherwise derived from the
/// compiled schema's declared `[validation]` limits (#379), an undeclared
/// depth being `DEFAULT_MAX_QUERY_DEPTH`. Always present: a bound binds on
/// every transport that reaches the executor, not just the `/graphql` stage.
pub(super) gate1: crate::security::QueryValidator,
/// Parsed GraphQL AST cache, keyed by xxHash64 of the query string.
pub(super) parse_cache: MokaCache<u64, Arc<(QueryType, Option<ParsedQuery>)>>,
/// Projected introspection responses, keyed by a hash of the normalised
/// selection set (#F7).
///
/// `__schema` must follow the client's selection set (GraphQL § 6.3), which
/// the pre-built response could not do. Projecting per request would trade
/// away the "zero-cost at runtime" property the canned response existed for
/// — except that projection is a **pure function of the selection set**, and
/// the space of introspection selection sets seen in the wild is tiny and
/// repetitive: `GraphiQL` sends one canonical query, Apollo sends one, each
/// codegen tool sends one, and they do not vary between page loads. Memoised
/// by shape, it is a table lookup again after the first hit.
///
/// Stores the projected `Value` behind an `Arc` so a hit is an O(1)
/// ref-count bump, matching how the unprojected response was served.
pub(super) introspection_projections: MokaCache<u64, Arc<serde_json::Value>>,
/// Optional executor-level response cache.
pub(super) response_cache: Option<Arc<crate::cache::ResponseCache>>,
}
impl ExecutorContext {
/// The write handle, or the refusal that names both gates.
///
/// The single adjudication of "may this executor write?". Both the document path
/// (`execute_mutation_query`) and the five typed write entries resolve it here, so
/// there is one decision rather than one per entry point — and the entries that
/// used to have no check at all cannot regain that state, because they need the
/// return value to dispatch.
///
/// # Errors
///
/// [`FraiseQLError::Validation`] naming the mutation when the executor was built
/// read-only.
pub(super) fn writer(&self, mutation_name: &str) -> Result<&dyn crate::backend::Writer> {
self.writer.as_deref().ok_or_else(|| crate::error::FraiseQLError::Validation {
message: format!(
"Mutation '{mutation_name}' cannot be executed: the configured database \
adapter is read-only, so this executor runs no mutations. A write-capable \
adapter implements `Writer` \
(`execute_write`), and the executor is built with `Executor::new` or \
`Executor::with_config`. `PostgresAdapter` does; `FraiseWireAdapter` \
deliberately does not."
),
path: None,
})
}
/// Return current connection pool metrics.
pub(super) fn pool_metrics(&self) -> PoolMetrics {
self.adapter.pool_metrics()
}
/// Which backend this executor is bound to.
pub(super) fn database_type(&self) -> DatabaseType {
self.adapter.database_type()
}
/// Whether the backend is reachable.
pub(super) async fn health_check(&self) -> Result<()> {
self.adapter.health_check().await
}
/// The slowest `limit` statements the backend is willing to report.
pub(super) async fn query_stats(&self, limit: u32) -> Result<Vec<QueryStatEntry>> {
self.adapter.query_stats(limit).await
}
/// One statement's stats by backend-assigned id.
pub(super) async fn query_stats_by_id(&self, id: &str) -> Result<Option<QueryStatEntry>> {
self.adapter.query_stats_by_id(id).await
}
/// Discard the backend's accumulated statement statistics.
pub(super) async fn reset_query_stats(&self) -> Result<()> {
self.adapter.reset_query_stats().await
}
/// Adapter-level result-cache counters, or `None` when no cache is active.
pub(super) fn result_cache_stats(&self) -> Option<ResultCacheStats> {
self.adapter.result_cache_stats()
}
/// Evict every entry from the adapter-level result cache.
pub(super) async fn clear_result_cache(&self) -> Result<Option<usize>> {
self.adapter.clear_result_cache().await
}
/// Evict adapter-level result-cache entries derived from the given views.
pub(super) async fn invalidate_views(&self, views: &[ViewName]) -> Result<u64> {
self.adapter.invalidate_views(views).await
}
/// The backend's plan for a statement, as its own JSON shape.
pub(super) async fn explain_query(
&self,
sql: &str,
params: &[serde_json::Value],
) -> Result<serde_json::Value> {
self.adapter.explain_query(sql, params).await
}
/// Tell the backend the compiled schema changed, so it can drop anything it
/// derived from the old one.
pub(super) fn on_schema_reload(&self) {
self.adapter.on_schema_reload();
}
/// Execute one server-composed DDL statement.
///
/// Private to this module: the only public door onto it is
/// [`Executor::drop_tenant_schema`], which composes the statement itself.
pub(super) async fn execute_ddl(&self, ddl: &str) -> Result<()> {
self.adapter.execute_raw_query(ddl).await.map(|_| ())
}
/// Run the admin-SQL route's bounded statement.
pub(super) async fn execute_admin_sql(
&self,
request: &AdminSqlRequest,
) -> Result<AdminSqlOutcome> {
self.adapter.execute_admin_sql(request).await
}
}