@remit/imap-worker 0.0.71 → 0.0.72
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@remit/imap-worker",
|
|
3
|
-
"version": "0.0.
|
|
3
|
+
"version": "0.0.72",
|
|
4
4
|
"type": "module",
|
|
5
5
|
"main": "dist/index.js",
|
|
6
6
|
"exports": {
|
|
@@ -39,7 +39,6 @@
|
|
|
39
39
|
"@remit/logger-lambda": "*",
|
|
40
40
|
"@remit/mailbox-service": "*",
|
|
41
41
|
"@remit/search-index-worker": "*",
|
|
42
|
-
"@remit/search-service": "*",
|
|
43
42
|
"@remit/sqs-client": "*",
|
|
44
43
|
"@remit/backend": "*",
|
|
45
44
|
"@remit/domain-enums": "*",
|
|
@@ -11,6 +11,7 @@ import {
|
|
|
11
11
|
QuarantineService,
|
|
12
12
|
resolveExhaustedBodySyncFailures,
|
|
13
13
|
} from "@remit/mailbox-service";
|
|
14
|
+
import { buildFilterConfig } from "@remit/mailbox-service/filter-config";
|
|
14
15
|
import { attemptBudget } from "@remit/sqs-client/attempt-budget";
|
|
15
16
|
import { env } from "expect-env";
|
|
16
17
|
import { isAccountDeleted } from "../account-check.js";
|
|
@@ -20,7 +21,6 @@ import {
|
|
|
20
21
|
createConnectionScopeWithCredentials,
|
|
21
22
|
} from "../connection-scope.js";
|
|
22
23
|
import type { SyncMessageBodyEvent } from "../events.js";
|
|
23
|
-
import { buildFilterConfig } from "../filter-config.js";
|
|
24
24
|
import { isNotFoundError } from "../is-not-found.js";
|
|
25
25
|
import { withOAuthLifecycle } from "../with-oauth-lifecycle.js";
|
|
26
26
|
import { buildLifecycleDeps } from "../with-oauth-lifecycle-deps.js";
|
|
@@ -1,171 +0,0 @@
|
|
|
1
|
-
import assert from "node:assert/strict";
|
|
2
|
-
import { describe, it } from "node:test";
|
|
3
|
-
import type {
|
|
4
|
-
FilterItem,
|
|
5
|
-
IFilterAnchorRepository,
|
|
6
|
-
IFilterRepository,
|
|
7
|
-
IMessageLabelRepository,
|
|
8
|
-
} from "@remit/data-ports";
|
|
9
|
-
import { FilterMatchOperator, FilterState } from "@remit/domain-enums";
|
|
10
|
-
import {
|
|
11
|
-
type FilterMessage,
|
|
12
|
-
FilterPipeline,
|
|
13
|
-
type MessageEmbedder,
|
|
14
|
-
NO_ACTION,
|
|
15
|
-
type PlacementMoveService,
|
|
16
|
-
} from "@remit/mailbox-service";
|
|
17
|
-
import { buildFilterConfig, type FilterConfigDeps } from "./filter-config.js";
|
|
18
|
-
import { _resetMessageEmbedderForTest } from "./message-embedder.js";
|
|
19
|
-
|
|
20
|
-
const anchorOnlyFilter = (destinationMailboxId: string): FilterItem =>
|
|
21
|
-
({
|
|
22
|
-
filterId: "flt-semantic",
|
|
23
|
-
accountConfigId: "cfg-1",
|
|
24
|
-
name: "receipts",
|
|
25
|
-
scope: "Standing",
|
|
26
|
-
state: FilterState.Active,
|
|
27
|
-
hasAnchor: true,
|
|
28
|
-
ruleChangedAt: 1,
|
|
29
|
-
matchOperator: FilterMatchOperator.Or,
|
|
30
|
-
literalClauses: [],
|
|
31
|
-
actionLabelId: NO_ACTION,
|
|
32
|
-
actionMailboxId: destinationMailboxId,
|
|
33
|
-
createdAt: 1,
|
|
34
|
-
updatedAt: 1,
|
|
35
|
-
}) as unknown as FilterItem;
|
|
36
|
-
|
|
37
|
-
const CURRENT_MODEL = "test-model@3";
|
|
38
|
-
|
|
39
|
-
const repos = (filter: FilterItem): FilterConfigDeps => ({
|
|
40
|
-
filterService: {
|
|
41
|
-
listByAccountAndState: async () => [filter],
|
|
42
|
-
refreshExpiry: async (f: FilterItem) => f,
|
|
43
|
-
} as unknown as IFilterRepository,
|
|
44
|
-
filterAnchorService: {
|
|
45
|
-
get: async () => ({
|
|
46
|
-
anchorEmbedding: [1, 0, 0],
|
|
47
|
-
anchorEmbeddingId: CURRENT_MODEL,
|
|
48
|
-
}),
|
|
49
|
-
} as unknown as IFilterAnchorRepository,
|
|
50
|
-
messageLabelService: {} as unknown as IMessageLabelRepository,
|
|
51
|
-
placementMoveService: {} as unknown as PlacementMoveService,
|
|
52
|
-
});
|
|
53
|
-
|
|
54
|
-
const message: FilterMessage = {
|
|
55
|
-
from: "billing@stripe.com",
|
|
56
|
-
fromName: "Stripe",
|
|
57
|
-
subject: "Your receipt",
|
|
58
|
-
text: "Thanks for your payment",
|
|
59
|
-
listId: "",
|
|
60
|
-
};
|
|
61
|
-
|
|
62
|
-
describe("buildFilterConfig", () => {
|
|
63
|
-
it("returns undefined without a placement mover — no move path, filters stay off", () => {
|
|
64
|
-
const deps = repos(anchorOnlyFilter("mbx-archive"));
|
|
65
|
-
deps.placementMoveService = undefined;
|
|
66
|
-
|
|
67
|
-
assert.equal(buildFilterConfig(deps), undefined);
|
|
68
|
-
});
|
|
69
|
-
|
|
70
|
-
it("wires an embedder into the config", () => {
|
|
71
|
-
const config = buildFilterConfig(repos(anchorOnlyFilter("mbx-archive")));
|
|
72
|
-
|
|
73
|
-
assert.ok(config);
|
|
74
|
-
assert.ok(config.embedder);
|
|
75
|
-
});
|
|
76
|
-
|
|
77
|
-
it("lights up a semantic anchor-only filter on the body-sync pass", async () => {
|
|
78
|
-
let embedCalls = 0;
|
|
79
|
-
const embedder: MessageEmbedder = {
|
|
80
|
-
embed: async () => {
|
|
81
|
-
embedCalls += 1;
|
|
82
|
-
return [1, 0, 0];
|
|
83
|
-
},
|
|
84
|
-
embeddingId: CURRENT_MODEL,
|
|
85
|
-
};
|
|
86
|
-
const config = buildFilterConfig(
|
|
87
|
-
repos(anchorOnlyFilter("mbx-archive")),
|
|
88
|
-
embedder,
|
|
89
|
-
);
|
|
90
|
-
assert.ok(config);
|
|
91
|
-
|
|
92
|
-
const decision = await new FilterPipeline(config, {
|
|
93
|
-
info: () => {},
|
|
94
|
-
}).evaluate("cfg-1", "m-1", message);
|
|
95
|
-
|
|
96
|
-
assert.deepEqual(decision.move, {
|
|
97
|
-
destinationMailboxId: "mbx-archive",
|
|
98
|
-
filterId: "flt-semantic",
|
|
99
|
-
});
|
|
100
|
-
assert.equal(embedCalls, 1);
|
|
101
|
-
});
|
|
102
|
-
|
|
103
|
-
it("does not match when the message embedding diverges from the anchor", async () => {
|
|
104
|
-
const embedder: MessageEmbedder = {
|
|
105
|
-
embed: async () => [0, 1, 0],
|
|
106
|
-
embeddingId: CURRENT_MODEL,
|
|
107
|
-
};
|
|
108
|
-
const config = buildFilterConfig(
|
|
109
|
-
repos(anchorOnlyFilter("mbx-archive")),
|
|
110
|
-
embedder,
|
|
111
|
-
);
|
|
112
|
-
assert.ok(config);
|
|
113
|
-
|
|
114
|
-
const decision = await new FilterPipeline(config, {
|
|
115
|
-
info: () => {},
|
|
116
|
-
}).evaluate("cfg-1", "m-2", message);
|
|
117
|
-
|
|
118
|
-
assert.equal(decision.move, undefined);
|
|
119
|
-
});
|
|
120
|
-
});
|
|
121
|
-
|
|
122
|
-
// An instance with semantic search off (issue #1068). `off` selects an embedder
|
|
123
|
-
// that throws by design, and the throw lands in the pipeline's per-filter catch:
|
|
124
|
-
// one error-level `filter_anchor_match_failed` for every semantic filter on
|
|
125
|
-
// every message it syncs. The pipeline already has a designed skip for a
|
|
126
|
-
// missing embedder, so the config passes none.
|
|
127
|
-
describe("buildFilterConfig with semantic search off", () => {
|
|
128
|
-
const withProviderOff = async (body: () => Promise<void>): Promise<void> => {
|
|
129
|
-
const previous = process.env.SEARCH_EMBEDDING_PROVIDER;
|
|
130
|
-
process.env.SEARCH_EMBEDDING_PROVIDER = "off";
|
|
131
|
-
_resetMessageEmbedderForTest();
|
|
132
|
-
try {
|
|
133
|
-
await body();
|
|
134
|
-
} finally {
|
|
135
|
-
if (previous === undefined) {
|
|
136
|
-
delete process.env.SEARCH_EMBEDDING_PROVIDER;
|
|
137
|
-
} else {
|
|
138
|
-
process.env.SEARCH_EMBEDDING_PROVIDER = previous;
|
|
139
|
-
}
|
|
140
|
-
_resetMessageEmbedderForTest();
|
|
141
|
-
}
|
|
142
|
-
};
|
|
143
|
-
|
|
144
|
-
it("wires no embedder", async () => {
|
|
145
|
-
await withProviderOff(async () => {
|
|
146
|
-
const config = buildFilterConfig(repos(anchorOnlyFilter("mbx-archive")));
|
|
147
|
-
|
|
148
|
-
assert.ok(config);
|
|
149
|
-
assert.equal(config.embedder, undefined);
|
|
150
|
-
});
|
|
151
|
-
});
|
|
152
|
-
|
|
153
|
-
it("skips the semantic filter without logging an error", async () => {
|
|
154
|
-
await withProviderOff(async () => {
|
|
155
|
-
const config = buildFilterConfig(repos(anchorOnlyFilter("mbx-archive")));
|
|
156
|
-
assert.ok(config);
|
|
157
|
-
|
|
158
|
-
const errors: unknown[] = [];
|
|
159
|
-
const decision = await new FilterPipeline(config, {
|
|
160
|
-
info: () => {},
|
|
161
|
-
debug: () => {},
|
|
162
|
-
warn: () => {},
|
|
163
|
-
error: (fields) => errors.push(fields.alert),
|
|
164
|
-
}).evaluate("cfg-1", "m-off", message);
|
|
165
|
-
|
|
166
|
-
assert.deepEqual(decision.labels, []);
|
|
167
|
-
assert.equal(decision.move, undefined);
|
|
168
|
-
assert.deepEqual(errors, []);
|
|
169
|
-
});
|
|
170
|
-
});
|
|
171
|
-
});
|
package/src/filter-config.ts
DELETED
|
@@ -1,63 +0,0 @@
|
|
|
1
|
-
import type {
|
|
2
|
-
IFilterAnchorRepository,
|
|
3
|
-
IFilterRepository,
|
|
4
|
-
IMessageLabelRepository,
|
|
5
|
-
} from "@remit/data-ports";
|
|
6
|
-
import type {
|
|
7
|
-
FilterConfig,
|
|
8
|
-
MessageEmbedder,
|
|
9
|
-
PlacementMoveService,
|
|
10
|
-
} from "@remit/mailbox-service";
|
|
11
|
-
import {
|
|
12
|
-
EMBEDDING_PROVIDER_OFF,
|
|
13
|
-
readEmbeddingProviderFromEnv,
|
|
14
|
-
} from "@remit/search-service/from-env";
|
|
15
|
-
import { getMessageEmbedder } from "./message-embedder.js";
|
|
16
|
-
|
|
17
|
-
export interface FilterConfigDeps {
|
|
18
|
-
filterService: IFilterRepository;
|
|
19
|
-
filterAnchorService: IFilterAnchorRepository;
|
|
20
|
-
messageLabelService: IMessageLabelRepository;
|
|
21
|
-
placementMoveService?: PlacementMoveService;
|
|
22
|
-
}
|
|
23
|
-
|
|
24
|
-
/**
|
|
25
|
-
* The env-selected embedder, or none on an instance with semantic search off.
|
|
26
|
-
*
|
|
27
|
-
* `off` embeds nothing by design, and its embedder says so by throwing
|
|
28
|
-
* (EmbeddingDisabledError). The pipeline has a designed skip for a *missing*
|
|
29
|
-
* embedder — semantic filters are passed over and logged at debug — but a
|
|
30
|
-
* present embedder that throws lands in the per-filter catch instead, which is
|
|
31
|
-
* an error-level `filter_anchor_match_failed` for every semantic filter on
|
|
32
|
-
* every synced message. Same outcome, a log nobody can read.
|
|
33
|
-
*/
|
|
34
|
-
const embedderFromEnv = (): MessageEmbedder | undefined =>
|
|
35
|
-
readEmbeddingProviderFromEnv() === EMBEDDING_PROVIDER_OFF
|
|
36
|
-
? undefined
|
|
37
|
-
: getMessageEmbedder();
|
|
38
|
-
|
|
39
|
-
/**
|
|
40
|
-
* Assemble the index-time filter config the body-sync pass runs (RFC 034). The
|
|
41
|
-
* embedder is provisioned from env exactly as the backend read-path and
|
|
42
|
-
* search-index worker provision theirs, so a semantic (anchor-only) filter is
|
|
43
|
-
* evaluated on incoming mail instead of silently skipped.
|
|
44
|
-
*
|
|
45
|
-
* Absent the placement mover there is no move path, so filters stay off — a
|
|
46
|
-
* matched filter's actions reuse the same enqueue plumbing the placement mover
|
|
47
|
-
* owns.
|
|
48
|
-
*/
|
|
49
|
-
export const buildFilterConfig = (
|
|
50
|
-
deps: FilterConfigDeps,
|
|
51
|
-
embedder?: MessageEmbedder,
|
|
52
|
-
): FilterConfig | undefined => {
|
|
53
|
-
const { placementMoveService } = deps;
|
|
54
|
-
if (!placementMoveService) return undefined;
|
|
55
|
-
const resolved = embedder ?? embedderFromEnv();
|
|
56
|
-
return {
|
|
57
|
-
filterService: deps.filterService,
|
|
58
|
-
filterAnchorService: deps.filterAnchorService,
|
|
59
|
-
messageLabelService: deps.messageLabelService,
|
|
60
|
-
placementMoveService,
|
|
61
|
-
...(resolved ? { embedder: resolved } : {}),
|
|
62
|
-
};
|
|
63
|
-
};
|
|
@@ -1,39 +0,0 @@
|
|
|
1
|
-
import assert from "node:assert/strict";
|
|
2
|
-
import { afterEach, describe, it } from "node:test";
|
|
3
|
-
import {
|
|
4
|
-
_resetMessageEmbedderForTest,
|
|
5
|
-
getMessageEmbedder,
|
|
6
|
-
} from "./message-embedder.js";
|
|
7
|
-
|
|
8
|
-
describe("getMessageEmbedder", () => {
|
|
9
|
-
afterEach(() => {
|
|
10
|
-
_resetMessageEmbedderForTest();
|
|
11
|
-
});
|
|
12
|
-
|
|
13
|
-
it("adapts the batch embedding service into a single-text message embedder", async () => {
|
|
14
|
-
const vector = await getMessageEmbedder().embed("invoice from stripe");
|
|
15
|
-
|
|
16
|
-
assert.ok(Array.isArray(vector));
|
|
17
|
-
assert.ok(vector.length > 0);
|
|
18
|
-
assert.ok(vector.every((value) => typeof value === "number"));
|
|
19
|
-
});
|
|
20
|
-
|
|
21
|
-
it("embeds the same text deterministically", async () => {
|
|
22
|
-
const embedder = getMessageEmbedder();
|
|
23
|
-
const a = await embedder.embed("your monthly receipt");
|
|
24
|
-
const b = await embedder.embed("your monthly receipt");
|
|
25
|
-
|
|
26
|
-
assert.deepEqual(a, b);
|
|
27
|
-
});
|
|
28
|
-
|
|
29
|
-
it("memoizes the embedder across calls", () => {
|
|
30
|
-
assert.strictEqual(getMessageEmbedder(), getMessageEmbedder());
|
|
31
|
-
});
|
|
32
|
-
|
|
33
|
-
it("rebuilds after a reset", () => {
|
|
34
|
-
const first = getMessageEmbedder();
|
|
35
|
-
_resetMessageEmbedderForTest();
|
|
36
|
-
|
|
37
|
-
assert.notStrictEqual(getMessageEmbedder(), first);
|
|
38
|
-
});
|
|
39
|
-
});
|
package/src/message-embedder.ts
DELETED
|
@@ -1,33 +0,0 @@
|
|
|
1
|
-
import type { MessageEmbedder } from "@remit/mailbox-service";
|
|
2
|
-
import { buildEmbeddingServiceFromEnv } from "@remit/search-service/from-env";
|
|
3
|
-
|
|
4
|
-
/**
|
|
5
|
-
* Adapt the batch {@link EmbeddingService} the rest of the stack composes from
|
|
6
|
-
* env into the single-text {@link MessageEmbedder} the filter pipeline needs: one
|
|
7
|
-
* incoming message becomes one message-level vector to compare against a filter's
|
|
8
|
-
* persisted anchor (RFC 034 Decision 2.1/2.3).
|
|
9
|
-
*
|
|
10
|
-
* The embedder is selected by the same `SEARCH_EMBEDDING_*` env the backend and
|
|
11
|
-
* search-index worker read, so the worker embeds under the identical model the
|
|
12
|
-
* anchors were built with. The instance is memoized across warm invocations so a
|
|
13
|
-
* local Transformers.js model loads once per container rather than per event.
|
|
14
|
-
*/
|
|
15
|
-
let cached: MessageEmbedder | undefined;
|
|
16
|
-
|
|
17
|
-
export const getMessageEmbedder = (): MessageEmbedder => {
|
|
18
|
-
if (cached) return cached;
|
|
19
|
-
const service = buildEmbeddingServiceFromEnv();
|
|
20
|
-
cached = {
|
|
21
|
-
embed: async (text: string): Promise<number[]> => {
|
|
22
|
-
const [vector] = await service.embed([text]);
|
|
23
|
-
return vector;
|
|
24
|
-
},
|
|
25
|
-
embeddingId: service.embeddingId,
|
|
26
|
-
};
|
|
27
|
-
return cached;
|
|
28
|
-
};
|
|
29
|
-
|
|
30
|
-
/** Reset the singleton — test use only. */
|
|
31
|
-
export const _resetMessageEmbedderForTest = (): void => {
|
|
32
|
-
cached = undefined;
|
|
33
|
-
};
|