cf-mini-chat 0.1.29

Mini-chat module: multi-tenant AI chat
Documentation 
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
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
250
251

use modkit_macros::domain_model;

/// Output port for recording domain-level metrics.
///
/// Implementations live in `infra/metrics.rs` (OpenTelemetry instruments).
/// Domain services depend only on this trait — no knowledge of `OTel`.
///
/// All 26 P0+P1 metrics are exposed here. Deferred (P2/P3) metrics are
/// declared as `OTel` instruments in the infra layer but NOT on this trait
/// until their feature is ready.
///
/// ## Naming convention
///
/// Instrument names intentionally omit the `_total` suffix that appears in
/// Prometheus metric names. The OpenTelemetry SDK (via
/// `opentelemetry-prometheus`) automatically appends `_total` to all
/// `Counter` instruments during Prometheus export, so including it in the
/// instrument name would produce a doubled `_total_total` suffix.
#[allow(dead_code)]
pub trait MiniChatMetricsPort: Send + Sync {
    // ── P0: Streaming & UX Health (8 metrics) ──────────────────────────
    //
    // ## Stream outcome counter invariant
    //
    // Terminal states are **disjoint**:
    //   `stream_started = stream_completed + stream_failed`
    //
    // `stream_incomplete` is a **diagnostic sub-counter** of `completed`,
    // not a separate terminal state:
    //   `stream_incomplete ⊆ stream_completed`
    //
    // An incomplete stream therefore increments *both*
    // `record_stream_incomplete` and `record_stream_completed`.
    // Dashboard authors: do NOT sum incomplete + completed — incomplete
    // is already included in the completed total.

    /// `{prefix}_stream_started_total` — counter
    fn record_stream_started(&self, provider: &str, model: &str);

    /// `{prefix}_stream_completed_total` — counter
    ///
    /// Covers both `response.completed` and `response.incomplete` outcomes.
    /// See the stream outcome invariant note above.
    fn record_stream_completed(&self, provider: &str, model: &str);

    /// `{prefix}_stream_failed_total` — counter
    fn record_stream_failed(&self, provider: &str, model: &str, error_code: &str);

    /// `{prefix}_stream_disconnected_total` — counter
    /// `stage`: `before_first_token`, `mid_stream`, `after_done`
    fn record_stream_disconnected(&self, stage: &str);

    /// `{prefix}_active_streams` — gauge increment/decrement
    fn increment_active_streams(&self);
    fn decrement_active_streams(&self);

    /// `{prefix}_ttft_provider_ms` — histogram (time-to-first-token from provider)
    fn record_ttft_provider_ms(&self, provider: &str, model: &str, ms: f64);

    /// `{prefix}_ttft_overhead_ms` — histogram (additional overhead beyond provider TTFT)
    fn record_ttft_overhead_ms(&self, provider: &str, model: &str, ms: f64);

    /// `{prefix}_stream_total_latency_ms` — histogram
    fn record_stream_total_latency_ms(&self, provider: &str, model: &str, ms: f64);

    // ── P0: Turn Mutations (2 metrics) ─────────────────────────────────

    /// `{prefix}_turn_mutation_total` — counter
    /// `op`: `retry`, `edit`, `delete`
    /// `result`: `ok`, `not_latest`, `invalid_state`, `forbidden`
    fn record_turn_mutation(&self, op: &str, result: &str);

    /// `{prefix}_turn_mutation_latency_ms` — histogram
    fn record_turn_mutation_latency_ms(&self, op: &str, ms: f64);

    // ── P0: Audit Emission Health (2 metrics) ──────────────────────────

    /// `{prefix}_audit_emit_total` — counter
    /// `result`: `ok`, `retry`, `reject`
    fn record_audit_emit(&self, result: &str);

    /// `{prefix}_finalization_latency_ms` — histogram
    fn record_finalization_latency_ms(&self, ms: f64);

    // ── P1: Quota Enforcement (6 metrics) ──────────────────────────────

    /// `{prefix}_quota_preflight_total` — counter
    fn record_quota_preflight(&self, decision: &str, model: &str, tier: &str);

    /// `{prefix}_quota_reserve_total` — counter
    fn record_quota_reserve(&self, period: &str);

    /// `{prefix}_quota_commit_total` — counter
    fn record_quota_commit(&self, period: &str);

    /// `{prefix}_quota_overshoot_total` — counter
    fn record_quota_overshoot(&self, period: &str);

    /// `{prefix}_quota_estimated_tokens` — histogram
    fn record_quota_estimated_tokens(&self, tokens: f64);

    /// `{prefix}_quota_actual_tokens` — histogram
    fn record_quota_actual_tokens(&self, tokens: f64);

    // ── P1: Streaming Incomplete (1 metric) ────────────────────────────

    /// `{prefix}_stream_incomplete_total` — counter
    ///
    /// Diagnostic sub-counter of `stream_completed`. Always called
    /// **alongside** `record_stream_completed`, never instead of it.
    /// See the stream outcome invariant note in the P0 section above.
    fn record_stream_incomplete(&self, provider: &str, model: &str, reason: &str);

    // ── P1: Cancellation (4 metrics) ───────────────────────────────────

    /// `{prefix}_cancel_requested_total` — counter
    /// `trigger`: `user_stop`, `disconnect`, `timeout`
    fn record_cancel_requested(&self, trigger: &str);

    /// `{prefix}_cancel_effective_total` — counter
    fn record_cancel_effective(&self, trigger: &str);

    /// `{prefix}_time_to_abort_ms` — histogram
    fn record_time_to_abort_ms(&self, trigger: &str, ms: f64);

    /// `{prefix}_streams_aborted_total` — counter
    /// `trigger`: `client_disconnect`, `pod_crash`, `orphan_timeout`, `internal_abort`
    fn record_streams_aborted(&self, trigger: &str);

    // ── P1: Attachment Upload (3 metrics) ──────────────────────────────

    /// `{prefix}_attachment_upload_total` — counter
    /// `kind`: `document`, `image`
    /// `result`: `ok`, `file_too_large`, `unsupported_type`, `provider_error`
    fn record_attachment_upload(&self, kind: &str, result: &str);

    /// `{prefix}_attachment_upload_bytes` — histogram
    fn record_attachment_upload_bytes(&self, kind: &str, bytes: f64);

    /// `{prefix}_attachments_pending` — gauge increment/decrement
    fn increment_attachments_pending(&self);
    fn decrement_attachments_pending(&self);

    /// `{prefix}_image_inputs_per_turn` — histogram
    fn record_image_inputs_per_turn(&self, count: u32);

    // ── P1: Cleanup (5 metrics) ─────────────────────────────────────────

    /// `{prefix}_cleanup_completed` — counter
    /// `resource_type`: `file`, `vector_store`
    fn record_cleanup_completed(&self, resource_type: &str);

    /// `{prefix}_cleanup_failed` — counter
    /// `resource_type`: `file`
    fn record_cleanup_failed(&self, resource_type: &str);

    /// `{prefix}_cleanup_retry` — counter
    /// `resource_type`: `file`, `vector_store`; `reason`: free-form
    fn record_cleanup_retry(&self, resource_type: &str, reason: &str);

    /// `{prefix}_cleanup_backlog` — gauge
    /// `state`: `pending`, `failed`; `resource_type`: `file`
    fn record_cleanup_backlog(&self, state: &str, resource_type: &str, count: i64);

    /// `{prefix}_cleanup_vector_store_with_failed_attachments` — counter
    fn record_cleanup_vs_with_failed_attachments(&self);

    // ── P1: Orphan Watchdog (3 metrics) ─────────────────────────────────

    /// `{prefix}_orphan_detected` — counter
    /// `reason`: `stale_progress`
    fn record_orphan_detected(&self, reason: &str);

    /// `{prefix}_orphan_finalized` — counter
    /// `reason`: `stale_progress`
    fn record_orphan_finalized(&self, reason: &str);

    /// `{prefix}_orphan_scan_duration_seconds` — histogram
    fn record_orphan_scan_duration_seconds(&self, seconds: f64);

    // ── P1: Thread Summary Health (4 metrics) ──────────────────────────

    /// `{prefix}_thread_summary_trigger` — counter
    /// `result`: `scheduled`, `not_needed`
    fn record_thread_summary_trigger(&self, result: &str);

    /// `{prefix}_thread_summary_execution` — counter
    /// `result`: `success`, `provider_error`, `timeout`, `retry`
    fn record_thread_summary_execution(&self, result: &str);

    /// `{prefix}_thread_summary_cas_conflicts` — counter
    fn record_thread_summary_cas_conflict(&self);

    /// `{prefix}_summary_fallback` — counter
    fn record_summary_fallback(&self);

    // ── P2: Tool Call Counters (1 metric) ────────────────────────────

    /// `{prefix}_code_interpreter_calls` — counter
    fn record_code_interpreter_calls(&self, model: &str, count: u32);
}

/// No-op implementation for use in tests or when metrics are disabled.
#[allow(dead_code)]
#[domain_model]
pub struct NoopMetrics;

impl MiniChatMetricsPort for NoopMetrics {
    fn record_stream_started(&self, _: &str, _: &str) {}
    fn record_stream_completed(&self, _: &str, _: &str) {}
    fn record_stream_failed(&self, _: &str, _: &str, _: &str) {}
    fn record_stream_disconnected(&self, _: &str) {}
    fn increment_active_streams(&self) {}
    fn decrement_active_streams(&self) {}
    fn record_ttft_provider_ms(&self, _: &str, _: &str, _: f64) {}
    fn record_ttft_overhead_ms(&self, _: &str, _: &str, _: f64) {}
    fn record_stream_total_latency_ms(&self, _: &str, _: &str, _: f64) {}
    fn record_turn_mutation(&self, _: &str, _: &str) {}
    fn record_turn_mutation_latency_ms(&self, _: &str, _: f64) {}
    fn record_audit_emit(&self, _: &str) {}
    fn record_finalization_latency_ms(&self, _: f64) {}
    fn record_quota_preflight(&self, _: &str, _: &str, _: &str) {}
    fn record_quota_reserve(&self, _: &str) {}
    fn record_quota_commit(&self, _: &str) {}
    fn record_quota_overshoot(&self, _: &str) {}
    fn record_quota_estimated_tokens(&self, _: f64) {}
    fn record_quota_actual_tokens(&self, _: f64) {}
    fn record_stream_incomplete(&self, _: &str, _: &str, _: &str) {}
    fn record_cancel_requested(&self, _: &str) {}
    fn record_cancel_effective(&self, _: &str) {}
    fn record_time_to_abort_ms(&self, _: &str, _: f64) {}
    fn record_streams_aborted(&self, _: &str) {}
    fn record_attachment_upload(&self, _: &str, _: &str) {}
    fn record_attachment_upload_bytes(&self, _: &str, _: f64) {}
    fn increment_attachments_pending(&self) {}
    fn decrement_attachments_pending(&self) {}
    fn record_image_inputs_per_turn(&self, _: u32) {}
    fn record_code_interpreter_calls(&self, _: &str, _: u32) {}
    fn record_cleanup_completed(&self, _: &str) {}
    fn record_cleanup_failed(&self, _: &str) {}
    fn record_cleanup_retry(&self, _: &str, _: &str) {}
    fn record_cleanup_backlog(&self, _: &str, _: &str, _: i64) {}
    fn record_cleanup_vs_with_failed_attachments(&self) {}
    fn record_orphan_detected(&self, _: &str) {}
    fn record_orphan_finalized(&self, _: &str) {}
    fn record_orphan_scan_duration_seconds(&self, _: f64) {}
    fn record_thread_summary_trigger(&self, _: &str) {}
    fn record_thread_summary_execution(&self, _: &str) {}
    fn record_thread_summary_cas_conflict(&self) {}
    fn record_summary_fallback(&self) {}
}