@nylorun/runtime 0.20.0-beta → 0.21.1-beta
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/CHANGELOG.md +86 -0
- package/README.md +19 -12
- package/dist/adapters/execution/restate.js +26 -1
- package/dist/api/components.d.ts +155 -36
- package/dist/api/components.js +3 -3
- package/dist/api/http/app.js +2 -0
- package/dist/api/http/routes/mcp.d.ts +9 -0
- package/dist/api/http/routes/mcp.js +32 -0
- package/dist/api/http/routes/vaults.d.ts +1 -9
- package/dist/api/http/routes/vaults.js +3 -103
- package/dist/api/openapi.js +6 -3
- package/dist/artifacts/media-types.d.ts +2 -0
- package/dist/artifacts/media-types.js +5 -0
- package/dist/core/flow-host.d.ts +2 -3
- package/dist/core/flow-host.js +10 -7
- package/dist/execution/types.d.ts +13 -5
- package/dist/execution/types.js +13 -5
- package/dist/gates/http-tool.d.ts +7 -0
- package/dist/gates/http-tool.js +34 -3
- package/dist/gates/mcp-handler.d.ts +4 -1
- package/dist/gates/mcp-handler.js +64 -8
- package/dist/gates/tenant-vaults.d.ts +9 -15
- package/dist/gates/tenant-vaults.js +9 -17
- package/dist/gates/tool-client.js +36 -17
- package/dist/gates/tool-contract.d.ts +19 -3
- package/dist/gates/tool-contract.js +9 -3
- package/dist/gates/tool-gate.d.ts +2 -2
- package/dist/harness/calls.d.ts +2 -0
- package/dist/harness/calls.js +13 -3
- package/dist/harness-api/record.d.ts +7 -6
- package/dist/harness-api/record.js +99 -17
- package/dist/harness-api/server.js +35 -27
- package/dist/harness-api/start.js +2 -1
- package/dist/host/config-for.d.ts +0 -2
- package/dist/host/config-for.js +0 -1
- package/dist/host/create-host.d.ts +1 -1
- package/dist/host/create-host.js +2 -1
- package/dist/host/gates.js +0 -1
- package/dist/host/http.d.ts +0 -1
- package/dist/host/http.js +0 -6
- package/dist/host/main.js +13 -3
- package/dist/host/stack-config.d.ts +7 -8
- package/dist/host/stack-config.js +50 -92
- package/dist/index.d.ts +0 -1
- package/dist/keys/client.js +1 -9
- package/dist/keys/keys.d.ts +11 -13
- package/dist/keys/keys.js +5 -7
- package/dist/management-openapi.json +315 -345
- package/dist/mcp/connect.d.ts +159 -13
- package/dist/mcp/connect.js +538 -55
- package/dist/mcp/pool.d.ts +11 -2
- package/dist/mcp/pool.js +74 -27
- package/dist/mcp/preview.d.ts +35 -0
- package/dist/mcp/preview.js +215 -0
- package/dist/mcp/snapshot.d.ts +79 -5
- package/dist/mcp/snapshot.js +145 -27
- package/dist/model/catalog.d.ts +14 -0
- package/dist/model/catalog.js +14 -1
- package/dist/model/pi-model.js +36 -2
- package/dist/openapi.json +362 -123
- package/dist/redact.d.ts +19 -1
- package/dist/redact.js +42 -13
- package/dist/sandbox/session-sandbox.d.ts +7 -0
- package/dist/sandbox/session-sandbox.js +89 -28
- package/dist/store/postgres/drizzle/0016_mcp_oauth_removed.sql +5 -0
- package/dist/store/postgres/drizzle/0017_wake_outbox.sql +9 -0
- package/dist/store/postgres/drizzle/meta/_journal.json +14 -0
- package/dist/store/postgres/schema.d.ts +105 -261
- package/dist/store/postgres/schema.js +18 -36
- package/dist/store/postgres/store.js +54 -21
- package/dist/store/postgres/tenant.d.ts +14 -0
- package/dist/store/postgres/tenant.js +22 -2
- package/dist/store/types.d.ts +48 -13
- package/dist/tenant/advance.js +1 -4
- package/dist/tenant/artifact-tool.d.ts +9 -1
- package/dist/tenant/artifact-tool.js +113 -9
- package/dist/tenant/commands.js +5 -5
- package/dist/tenant/context.d.ts +15 -13
- package/dist/tenant/effects.d.ts +2 -2
- package/dist/tenant/effects.js +20 -5
- package/dist/tenant/ephemeral.d.ts +0 -6
- package/dist/tenant/ephemeral.js +0 -1
- package/dist/tenant/outbound.d.ts +14 -5
- package/dist/tenant/outbound.js +21 -9
- package/dist/tenant/runtime.js +11 -15
- package/dist/tenant/sessions.js +20 -3
- package/dist/tenant/store-pg.js +9 -0
- package/dist/tenant/sweep.d.ts +14 -0
- package/dist/tenant/sweep.js +69 -10
- package/dist/tenant/tool-results.d.ts +57 -0
- package/dist/tenant/tool-results.js +298 -0
- package/dist/tenant/tool-search.d.ts +43 -0
- package/dist/tenant/tool-search.js +100 -0
- package/dist/tenant/transcript.d.ts +108 -0
- package/dist/tenant/transcript.js +13 -0
- package/dist/tenant/types.d.ts +0 -14
- package/dist/vault/crypto.d.ts +0 -7
- package/dist/vault/crypto.js +0 -16
- package/dist/vault/headers.d.ts +17 -0
- package/dist/vault/headers.js +61 -0
- package/dist/vault/service.d.ts +11 -68
- package/dist/vault/service.js +83 -438
- package/dist/vault/sources.d.ts +20 -48
- package/dist/vault/sources.js +25 -189
- package/dist/version.d.ts +1 -1
- package/dist/version.js +1 -1
- package/package.json +3 -3
- package/dist/vault/oauth.d.ts +0 -55
- package/dist/vault/oauth.js +0 -186
package/dist/tenant/runtime.js
CHANGED
|
@@ -5,7 +5,6 @@ import { tenantModelGate } from "../gates/in-process.js";
|
|
|
5
5
|
import { createKekFile, readVaultKek } from "../vault/kek.js";
|
|
6
6
|
import { SigningKeys } from "./signing-keys.js";
|
|
7
7
|
import { VaultService } from "../vault/service.js";
|
|
8
|
-
import { CredentialSources } from "../vault/sources.js";
|
|
9
8
|
import { McpPool } from "../mcp/pool.js";
|
|
10
9
|
import { SandboxManager } from "../sandbox/manager.js";
|
|
11
10
|
import { storeSandboxRecords } from "../sandbox/records.js";
|
|
@@ -22,7 +21,6 @@ import { sweep } from "./sweep.js";
|
|
|
22
21
|
import { TenantWorkers, WORKER_ID, } from "./worker.js";
|
|
23
22
|
import { authorize } from "./effects.js";
|
|
24
23
|
import { inProcessToolGate } from "../gates/tool-gate.js";
|
|
25
|
-
import { guardedFetch } from "./outbound.js";
|
|
26
24
|
import { inProcessKeys } from "../keys/keys.js";
|
|
27
25
|
import { tenantApi } from "../api/http/app.js";
|
|
28
26
|
import { createFsBlobStore } from "../blob/index.js";
|
|
@@ -118,16 +116,7 @@ export class TenantRuntime {
|
|
|
118
116
|
const opened = store;
|
|
119
117
|
// A seeded Tenant setting wins over the Host default (A18).
|
|
120
118
|
const seededBackend = await opened.tx((t) => t.getSetting("sandbox.backend"));
|
|
121
|
-
const vault = new VaultService({
|
|
122
|
-
store: opened,
|
|
123
|
-
kek: ensureKek,
|
|
124
|
-
// OAuth refresh and connect follow the Host's address policy (F9 C2).
|
|
125
|
-
fetch: config.vaultFetch ?? guardedFetch(config.delivery ?? {}),
|
|
126
|
-
});
|
|
127
|
-
const credentials = new CredentialSources({
|
|
128
|
-
vault,
|
|
129
|
-
...(config.resolver ? { resolver: config.resolver } : {}),
|
|
130
|
-
});
|
|
119
|
+
const vault = new VaultService({ store: opened, kek: ensureKek });
|
|
131
120
|
// `ctx` is assigned below; these callbacks only run once the Tenant is open.
|
|
132
121
|
let ctx;
|
|
133
122
|
const toolGate = hooks.toolGate ??
|
|
@@ -137,7 +126,14 @@ export class TenantRuntime {
|
|
|
137
126
|
logger: config.logger,
|
|
138
127
|
});
|
|
139
128
|
const signingKeys = new SigningKeys({ tenantId: config.tenantId, kek: ensureKek });
|
|
140
|
-
const keys = hooks.keys ??
|
|
129
|
+
const keys = hooks.keys ??
|
|
130
|
+
inProcessKeys({
|
|
131
|
+
store: opened,
|
|
132
|
+
vault,
|
|
133
|
+
signingKeys,
|
|
134
|
+
kek: ensureKek,
|
|
135
|
+
policy: config.delivery ?? {},
|
|
136
|
+
});
|
|
141
137
|
const harnessMode = hooks.harness ?? "memory";
|
|
142
138
|
const inProcess = harnessMode === "memory" || harnessMode === "json";
|
|
143
139
|
const preference = seededBackend ?? config.sandbox.backend;
|
|
@@ -242,7 +238,6 @@ export class TenantRuntime {
|
|
|
242
238
|
reads: hooks.reads,
|
|
243
239
|
store: opened,
|
|
244
240
|
vault,
|
|
245
|
-
credentials,
|
|
246
241
|
...(mcp ? { mcp } : {}),
|
|
247
242
|
sandbox,
|
|
248
243
|
flowLimits,
|
|
@@ -266,8 +261,9 @@ export class TenantRuntime {
|
|
|
266
261
|
ownerLeaseMs: config.ownerLeaseMs ?? DEFAULT_OWNER_LEASE_MS,
|
|
267
262
|
wake: async (sessionId, wake) => {
|
|
268
263
|
if (ctx.closing || ctx.closed)
|
|
269
|
-
return;
|
|
264
|
+
return false;
|
|
270
265
|
await execution.wake(config.tenantId, sessionId, wake);
|
|
266
|
+
return true;
|
|
271
267
|
},
|
|
272
268
|
abortLocal: (sessionId, turnId) => abortLocal(ctx, sessionId, "cancel", turnId),
|
|
273
269
|
...(hooks.pods ? { pods: hooks.pods } : {}),
|
package/dist/tenant/sessions.js
CHANGED
|
@@ -10,7 +10,7 @@ import { aggregateWaits, isWorkflowManifest } from "../core/flow-host.js";
|
|
|
10
10
|
import { canonical } from "../store/canonical.js";
|
|
11
11
|
import { validateSandboxAttach } from "../core/sandbox-routes.js";
|
|
12
12
|
import { resolveSandbox } from "../sandbox/resolve.js";
|
|
13
|
-
import { withSandboxCapability } from "../sandbox/session-sandbox.js";
|
|
13
|
+
import { withPlatformTools, withSandboxCapability } from "../sandbox/session-sandbox.js";
|
|
14
14
|
import { declaredSandboxes, sandboxSpecOf, sessionSandboxSpec, } from "../sandbox/share.js";
|
|
15
15
|
import { effectiveSandboxConfig, readSandboxConfig } from "../sandbox/tenant-config.js";
|
|
16
16
|
import { checkPlacement, requirePods } from "../sandbox/placement.js";
|
|
@@ -156,12 +156,14 @@ export async function putSession(ctx, id, body, access, options = {}) {
|
|
|
156
156
|
checkPlacement(await readSandboxConfig(t), sandbox.kind);
|
|
157
157
|
if (sandbox.kind === "pod")
|
|
158
158
|
await requirePods(ctx.pods, `Sandbox ${sandbox.sandboxId}`);
|
|
159
|
+
const pinned = sandbox.manifest === undefined ? platformTools(definition) : sandbox;
|
|
159
160
|
const created = {
|
|
160
161
|
id,
|
|
161
162
|
agentId: body.agentId,
|
|
162
163
|
ownerUserId: body.ownerUserId,
|
|
163
|
-
manifest:
|
|
164
|
-
manifestHash:
|
|
164
|
+
manifest: pinned.manifest ?? definition.manifest,
|
|
165
|
+
manifestHash: pinned.manifestHash ?? definition.manifestHash,
|
|
166
|
+
definitionHash: definition.manifestHash,
|
|
165
167
|
implementationVersion: definition.implementationVersion,
|
|
166
168
|
info: body.info,
|
|
167
169
|
status: "idle",
|
|
@@ -232,6 +234,20 @@ async function sessionSandbox(t, body, definition, options) {
|
|
|
232
234
|
return fail(resolved.status, resolved.errors.join(" "));
|
|
233
235
|
return pin(definition, resolved.spec, resolved.source);
|
|
234
236
|
}
|
|
237
|
+
/**
|
|
238
|
+
* The manifest a session without a sandbox pins: its agents' `read_artifact` (R2b C11), when
|
|
239
|
+
* one has a remote MCP server or an HTTP tool; nothing otherwise, and for a workflow.
|
|
240
|
+
*/
|
|
241
|
+
function platformTools(definition) {
|
|
242
|
+
if (isWorkflowManifest(definition.manifest))
|
|
243
|
+
return {};
|
|
244
|
+
const pinned = withPlatformTools(definition.manifest);
|
|
245
|
+
if (pinned === undefined)
|
|
246
|
+
return {};
|
|
247
|
+
if (!pinned.ok)
|
|
248
|
+
return fail(400, pinned.message);
|
|
249
|
+
return { manifest: pinned.manifest, manifestHash: pinned.manifestHash };
|
|
250
|
+
}
|
|
235
251
|
async function pin(definition, spec, source) {
|
|
236
252
|
// A workflow's manifest stays as registered: its agents inherit the sandbox when they start.
|
|
237
253
|
if (isWorkflowManifest(definition.manifest))
|
|
@@ -269,6 +285,7 @@ export async function sessionView(t, s) {
|
|
|
269
285
|
agentId: s.agentId,
|
|
270
286
|
ownerUserId: s.ownerUserId,
|
|
271
287
|
manifestHash: s.manifestHash,
|
|
288
|
+
...(s.definitionHash === undefined ? {} : { definitionHash: s.definitionHash }),
|
|
272
289
|
implementationVersion: s.implementationVersion,
|
|
273
290
|
status: s.status,
|
|
274
291
|
activeTurnId: s.activeTurnId,
|
package/dist/tenant/store-pg.js
CHANGED
|
@@ -55,6 +55,15 @@ export function createPostgresTenantOpener(options) {
|
|
|
55
55
|
from: opened.migrated.from,
|
|
56
56
|
to: opened.migrated.to,
|
|
57
57
|
});
|
|
58
|
+
// Protocol 10 removed the vault's `oauth` credential type: the migration deleted them.
|
|
59
|
+
for (const removed of opened.removedCredentials)
|
|
60
|
+
logger?.warn("oauth_credential_removed", {
|
|
61
|
+
tenantId,
|
|
62
|
+
vaultId: removed.vaultId,
|
|
63
|
+
credentialId: removed.credentialId,
|
|
64
|
+
url: removed.url,
|
|
65
|
+
message: "MCP OAuth credentials are gone (protocol 10): store a key for this server in a vault, or reach it through a gateway",
|
|
66
|
+
});
|
|
58
67
|
let reads;
|
|
59
68
|
try {
|
|
60
69
|
reads = createPostgresReadStore(options.sql, tenantId);
|
package/dist/tenant/sweep.d.ts
CHANGED
|
@@ -1,5 +1,19 @@
|
|
|
1
1
|
import type { TenantContext } from "./context.js";
|
|
2
|
+
/**
|
|
3
|
+
* How old a wake outbox row is before the sweep delivers it: longer than a delivery after
|
|
4
|
+
* commit takes, so the sweep seldom sends a wake its commit is still sending. Sending one
|
|
5
|
+
* twice is harmless either way (the idempotency key).
|
|
6
|
+
*/
|
|
7
|
+
export declare const WAKE_GRACE_MS = 2000;
|
|
8
|
+
/** How often the orphan scan runs after the first pass of a Tenant opened on this process. */
|
|
9
|
+
export declare const ORPHAN_SCAN_MS = 60000;
|
|
2
10
|
export declare function sweep(ctx: TenantContext): Promise<void>;
|
|
11
|
+
/**
|
|
12
|
+
* Deliver the wake outbox rows written before `before` (oldest first, one batch) and delete
|
|
13
|
+
* the delivered ones. A row the Tenant declines (it is closing) stays. Returns the ids of the
|
|
14
|
+
* rows delivered.
|
|
15
|
+
*/
|
|
16
|
+
export declare function deliverPendingWakes(ctx: Pick<TenantContext, "store" | "wake">, before?: Date): Promise<string[]>;
|
|
3
17
|
/** Settle pending workflow `agent` effects whose linked turn already finished. */
|
|
4
18
|
export declare function reconcileLinkedAgents(ctx: Pick<TenantContext, "store" | "wake">): Promise<void>;
|
|
5
19
|
/** Wake `running`/`runnable` sessions without a live owner. Returns the ids woken. */
|
package/dist/tenant/sweep.js
CHANGED
|
@@ -1,21 +1,32 @@
|
|
|
1
1
|
/**
|
|
2
2
|
* The Tenant sweep (architecture §12.3, "Timers"): one self-re-arming pass per Tenant, armed
|
|
3
3
|
* with `DurableExecution.armSweep` when the Tenant opens. It replaces every in-process timer
|
|
4
|
-
* and startup scan the Tenant runtime used to keep
|
|
4
|
+
* and startup scan the Tenant runtime used to keep. It reconciles the Session Store with what
|
|
5
|
+
* Durable Session Execution was asked to do; it never decides when a session advances:
|
|
5
6
|
*
|
|
6
|
-
* 1. **
|
|
7
|
+
* 1. **Wakes.** Wake outbox rows (`Tx.wake`) are delivered again and deleted: a delivery
|
|
8
|
+
* after commit that failed, or that a crash between the commit and the send cut off. The
|
|
9
|
+
* first pass on this process takes every row, so a previous process's are not left
|
|
10
|
+
* waiting; later passes take the rows older than `WAKE_GRACE_MS`. Each row carries the
|
|
11
|
+
* idempotency key it was first sent with, so a wake the execution did accept before the
|
|
12
|
+
* crash, or one still being sent after its commit, causes no second advance.
|
|
13
|
+
* 2. **Linked agents.** Pending workflow `agent` effects whose linked turn already settled
|
|
7
14
|
* are completed and the workflow is woken (a crash between the two, or data written by an
|
|
8
|
-
* older Runtime).
|
|
9
|
-
*
|
|
10
|
-
*
|
|
11
|
-
*
|
|
12
|
-
*
|
|
15
|
+
* older Runtime).
|
|
16
|
+
* 3. **Orphaned sessions.** On this process's first pass for the Tenant, then at most every
|
|
17
|
+
* `ORPHAN_SCAN_MS`: `running` or `runnable` sessions with no owner, or an owner whose lease
|
|
18
|
+
* expired, are woken with reason `recover`. Neither a lost wake (step 1) nor a Worker that
|
|
19
|
+
* died mid-advance (the execution retries that advance, which takes over, §11.4) needs it:
|
|
20
|
+
* it is the backstop for what the execution itself lost or ended without an advance, such
|
|
21
|
+
* as an in-process execution's queue when its process stopped, an advance for a Tenant not
|
|
22
|
+
* open on its Worker, an invocation an operator killed, or Restate's state wiped (§17.11).
|
|
23
|
+
* 4. **Sandboxes.** Idle sandboxes are stopped, records of compute no longer held are marked
|
|
13
24
|
* stopped, and sandboxes whose session or sandbox resource is gone are removed, by the
|
|
14
25
|
* workspace capability (`harness-api/workspace.ts`: here, or in the harness that serves
|
|
15
26
|
* workspaces). Idle MCP connections of the in-process harness are closed.
|
|
16
|
-
*
|
|
27
|
+
* 5. **Control signals.** Signals on the control bus older than `SIGNAL_RETENTION_MS` are
|
|
17
28
|
* deleted: followers read back only the last two minutes (`store/postgres/control.ts`).
|
|
18
|
-
*
|
|
29
|
+
* 6. **Hooks.** Callbacks registered with `ctx.onSweep`.
|
|
19
30
|
*
|
|
20
31
|
* Every step runs one transaction per session it changes (a linked agent and its workflow
|
|
21
32
|
* share one, child first), so the sweep follows the lock order in `store/types.ts`. A failing
|
|
@@ -26,13 +37,41 @@ import { pendingAgentEffects, reconcilePendingAgentEffect, } from "../core/flow-
|
|
|
26
37
|
const BATCH = 100;
|
|
27
38
|
/** How long a control signal is kept after it was written. */
|
|
28
39
|
const SIGNAL_RETENTION_MS = 60 * 60 * 1000;
|
|
40
|
+
/**
|
|
41
|
+
* How old a wake outbox row is before the sweep delivers it: longer than a delivery after
|
|
42
|
+
* commit takes, so the sweep seldom sends a wake its commit is still sending. Sending one
|
|
43
|
+
* twice is harmless either way (the idempotency key).
|
|
44
|
+
*/
|
|
45
|
+
export const WAKE_GRACE_MS = 2000;
|
|
46
|
+
/** How often the orphan scan runs after the first pass of a Tenant opened on this process. */
|
|
47
|
+
export const ORPHAN_SCAN_MS = 60_000;
|
|
48
|
+
/** Each open Tenant's sweep state on this process: when its orphan scan last ran. */
|
|
49
|
+
const passes = new WeakMap();
|
|
29
50
|
export async function sweep(ctx) {
|
|
30
51
|
if (ctx.closing || ctx.closed)
|
|
31
52
|
return;
|
|
32
53
|
const now = new Date();
|
|
54
|
+
const first = !passes.has(ctx);
|
|
55
|
+
const pass = passes.get(ctx) ?? {};
|
|
56
|
+
passes.set(ctx, pass);
|
|
57
|
+
const scanOrphans = pass.orphansAt === undefined || now.getTime() - pass.orphansAt >= ORPHAN_SCAN_MS;
|
|
33
58
|
const steps = [
|
|
59
|
+
[
|
|
60
|
+
"wakes",
|
|
61
|
+
() => deliverPendingWakes(ctx, first ? now : new Date(now.getTime() - WAKE_GRACE_MS)),
|
|
62
|
+
],
|
|
34
63
|
["linked", () => reconcileLinkedAgents(ctx)],
|
|
35
|
-
|
|
64
|
+
...(scanOrphans
|
|
65
|
+
? [
|
|
66
|
+
[
|
|
67
|
+
"orphans",
|
|
68
|
+
async () => {
|
|
69
|
+
await wakeOrphanedSessions(ctx, now);
|
|
70
|
+
pass.orphansAt = now.getTime();
|
|
71
|
+
},
|
|
72
|
+
],
|
|
73
|
+
]
|
|
74
|
+
: []),
|
|
36
75
|
[
|
|
37
76
|
"sandboxes",
|
|
38
77
|
() => ctx.sandbox.sweep({
|
|
@@ -68,6 +107,26 @@ export async function sweep(ctx) {
|
|
|
68
107
|
if (failure)
|
|
69
108
|
throw failure.error;
|
|
70
109
|
}
|
|
110
|
+
/**
|
|
111
|
+
* Deliver the wake outbox rows written before `before` (oldest first, one batch) and delete
|
|
112
|
+
* the delivered ones. A row the Tenant declines (it is closing) stays. Returns the ids of the
|
|
113
|
+
* rows delivered.
|
|
114
|
+
*/
|
|
115
|
+
export async function deliverPendingWakes(ctx, before = new Date()) {
|
|
116
|
+
const pending = await ctx.store.tx((t) => t.pendingWakes(before, BATCH));
|
|
117
|
+
const delivered = [];
|
|
118
|
+
try {
|
|
119
|
+
for (const row of pending)
|
|
120
|
+
if ((await ctx.wake(row.sessionId, row.wake)) !== false)
|
|
121
|
+
delivered.push(row.id);
|
|
122
|
+
}
|
|
123
|
+
finally {
|
|
124
|
+
// Delivered ones go even when a later one fails, so they are not sent again.
|
|
125
|
+
if (delivered.length > 0)
|
|
126
|
+
await ctx.store.tx((t) => t.deleteWakes(delivered));
|
|
127
|
+
}
|
|
128
|
+
return delivered;
|
|
129
|
+
}
|
|
71
130
|
/** Settle pending workflow `agent` effects whose linked turn already finished. */
|
|
72
131
|
export async function reconcileLinkedAgents(ctx) {
|
|
73
132
|
const { store } = ctx;
|
|
@@ -0,0 +1,57 @@
|
|
|
1
|
+
import type { HostEffect } from "@nylorun/harness/run";
|
|
2
|
+
import type { TenantContext } from "./context.js";
|
|
3
|
+
/** The most of one tool result that is recorded inline (Q22): 32 KiB, as JSON. */
|
|
4
|
+
export declare const TOOL_RESULT_INLINE_BYTES: number;
|
|
5
|
+
/** How much of a stored result's start, and of its end, the preview keeps. */
|
|
6
|
+
export declare const PREVIEW_HEAD_BYTES: number;
|
|
7
|
+
export declare const PREVIEW_TAIL_BYTES = 1024;
|
|
8
|
+
/**
|
|
9
|
+
* The most one step's results may hold together: a step's results are one transcript entry, so
|
|
10
|
+
* one event, which must stay well under S2's 1 MiB record.
|
|
11
|
+
*/
|
|
12
|
+
export declare const STEP_RESULTS_BUDGET_BYTES: number;
|
|
13
|
+
/** A result's cap once its step's budget is spent: a stub naming its artifact. */
|
|
14
|
+
export declare const TOOL_RESULT_STUB_BYTES = 512;
|
|
15
|
+
/** A result this small is never shaped, nor counted against its step's budget. */
|
|
16
|
+
export declare const TOOL_RESULT_SMALL_BYTES = 1024;
|
|
17
|
+
/** The call whose outcome is shaped, and the session its artifacts belong to. */
|
|
18
|
+
export interface ShapedCall {
|
|
19
|
+
readonly request: HostEffect;
|
|
20
|
+
readonly session: {
|
|
21
|
+
readonly id: string;
|
|
22
|
+
readonly ownerUserId: string;
|
|
23
|
+
};
|
|
24
|
+
/** A remote MCP tool's: its output may be the server's content parts. */
|
|
25
|
+
readonly mcp: boolean;
|
|
26
|
+
/** This result's cap (`resultLimit`). Default `TOOL_RESULT_INLINE_BYTES`. */
|
|
27
|
+
readonly limit?: number;
|
|
28
|
+
}
|
|
29
|
+
/** A result's cap when the other results of its step already took `used` bytes. */
|
|
30
|
+
export declare function resultLimit(used: number): number;
|
|
31
|
+
/** The JSON size of `value` in bytes: what it adds to an event or a prompt. */
|
|
32
|
+
export declare function jsonBytes(value: unknown): number;
|
|
33
|
+
/**
|
|
34
|
+
* What a tool outcome's result weighs in its step's transcript entry: a completed one's output,
|
|
35
|
+
* a failed one's message; 0 for anything else, such as a model call's outcome.
|
|
36
|
+
*/
|
|
37
|
+
export declare function resultBytes(value: unknown): number;
|
|
38
|
+
/**
|
|
39
|
+
* False for an outcome `shapeToolOutcome` returns as it is, whatever made it: a model call's, a
|
|
40
|
+
* small tool result with no parts. Spares a look at the effect for most outcomes.
|
|
41
|
+
*/
|
|
42
|
+
export declare function mayNeedShaping(value: unknown): boolean;
|
|
43
|
+
/**
|
|
44
|
+
* The outcome `value` of `call`, shaped to fit its cap (see the module comment). An outcome that
|
|
45
|
+
* fits and has no binary part comes back as it is.
|
|
46
|
+
*/
|
|
47
|
+
export declare function shapeToolOutcome(ctx: TenantContext, call: ShapedCall, value: unknown, signal: AbortSignal): Promise<unknown>;
|
|
48
|
+
/**
|
|
49
|
+
* The start and end of `text` in whole characters, `budget` bytes split four to one: by default
|
|
50
|
+
* the first `PREVIEW_HEAD_BYTES` and the last `PREVIEW_TAIL_BYTES`.
|
|
51
|
+
*/
|
|
52
|
+
export declare function previewOf(text: string, budget?: number): string;
|
|
53
|
+
/**
|
|
54
|
+
* The nearest character boundary of the UTF-8 `bytes` at or before (`down`) or at or after
|
|
55
|
+
* (`up`) byte `at`: never inside a character.
|
|
56
|
+
*/
|
|
57
|
+
export declare function utf8Boundary(bytes: Uint8Array, at: number, direction: "down" | "up"): number;
|
|
@@ -0,0 +1,298 @@
|
|
|
1
|
+
/**
|
|
2
|
+
* Tool results that fit (R2b C11, Q22–Q24): core shapes what an agent's remote MCP tool or HTTP
|
|
3
|
+
* tool returned before it records the outcome (`harness-api/record.ts`), so no journal row,
|
|
4
|
+
* event or prompt holds more of one result than the inline cap. The gate already refused an
|
|
5
|
+
* answer past 8 MiB (`mcp.too-large`, `http.too-large`).
|
|
6
|
+
*
|
|
7
|
+
* - An `image`, `audio` or blob `resource` part becomes a file artifact of the session, and the
|
|
8
|
+
* part says `{artifactId, version, contentType, size}`. An image also goes to the model as a
|
|
9
|
+
* file (`ToolOutcome.files`), which the model call shows to a model that reads images (Q23).
|
|
10
|
+
* - Text or JSON past `TOOL_RESULT_INLINE_BYTES` (32 KiB, Q22) becomes a file artifact, and the
|
|
11
|
+
* model gets `{truncated: true, artifactId, size, preview}`: the first 4 KiB and the last 1 KiB.
|
|
12
|
+
* In a result of several parts each part is shaped alone, the largest text first, until the
|
|
13
|
+
* whole fits; one that still does not is stored whole.
|
|
14
|
+
* - A `resource_link` stays a link, never fetched, and a text resource that fits stays inline.
|
|
15
|
+
* - A failed outcome's message is cut to its start and end.
|
|
16
|
+
* - When an artifact cannot be stored (the Tenant's artifact limits, an Object store that
|
|
17
|
+
* fails), the preview stays and `dropped` says why the rest is gone: a result never passes
|
|
18
|
+
* the cap. Only an abort throws.
|
|
19
|
+
* - **One step's results share a budget** (`STEP_RESULTS_BUDGET_BYTES`, 256 KiB), since the
|
|
20
|
+
* transcript keeps a step's results in one entry, one event: the record seam counts what the
|
|
21
|
+
* step's other results took, under the session lock, and gives this one the rest as its cap
|
|
22
|
+
* (`resultLimit`). Once the budget is spent, a result past `TOOL_RESULT_SMALL_BYTES` becomes a
|
|
23
|
+
* stub naming its artifact, its preview cut to fit `TOOL_RESULT_STUB_BYTES`.
|
|
24
|
+
*
|
|
25
|
+
* The model reads a stored result with `read_artifact` (`artifact-tool.ts`). Artifacts are the
|
|
26
|
+
* session's, saved by our engine (`source: "engine"`), with the turn and the tool call on their
|
|
27
|
+
* `artifact.created` event.
|
|
28
|
+
*/
|
|
29
|
+
import { READ_ARTIFACT_MAX_BYTES } from "@nylorun/core/define";
|
|
30
|
+
import { MODEL_FILE_MAX_BYTES, extensionFor, isModelImage, isText, normalizeContentType, } from "../artifacts/media-types.js";
|
|
31
|
+
import { uploadArtifact } from "../artifacts/service.js";
|
|
32
|
+
import { DEFAULT_CONTENT_TYPE } from "../blob/index.js";
|
|
33
|
+
import { HttpError } from "./http.js";
|
|
34
|
+
import { toolIds } from "./transcript.js";
|
|
35
|
+
/** The most of one tool result that is recorded inline (Q22): 32 KiB, as JSON. */
|
|
36
|
+
export const TOOL_RESULT_INLINE_BYTES = READ_ARTIFACT_MAX_BYTES;
|
|
37
|
+
/** How much of a stored result's start, and of its end, the preview keeps. */
|
|
38
|
+
export const PREVIEW_HEAD_BYTES = 4 * 1024;
|
|
39
|
+
export const PREVIEW_TAIL_BYTES = 1024;
|
|
40
|
+
/**
|
|
41
|
+
* The most one step's results may hold together: a step's results are one transcript entry, so
|
|
42
|
+
* one event, which must stay well under S2's 1 MiB record.
|
|
43
|
+
*/
|
|
44
|
+
export const STEP_RESULTS_BUDGET_BYTES = 256 * 1024;
|
|
45
|
+
/** A result's cap once its step's budget is spent: a stub naming its artifact. */
|
|
46
|
+
export const TOOL_RESULT_STUB_BYTES = 512;
|
|
47
|
+
/** A result this small is never shaped, nor counted against its step's budget. */
|
|
48
|
+
export const TOOL_RESULT_SMALL_BYTES = 1024;
|
|
49
|
+
/** A result's cap when the other results of its step already took `used` bytes. */
|
|
50
|
+
export function resultLimit(used) {
|
|
51
|
+
return Math.max(TOOL_RESULT_STUB_BYTES, Math.min(TOOL_RESULT_INLINE_BYTES, STEP_RESULTS_BUDGET_BYTES - used));
|
|
52
|
+
}
|
|
53
|
+
/** The JSON size of `value` in bytes: what it adds to an event or a prompt. */
|
|
54
|
+
export function jsonBytes(value) {
|
|
55
|
+
return Buffer.byteLength(JSON.stringify(value) ?? "null");
|
|
56
|
+
}
|
|
57
|
+
/**
|
|
58
|
+
* What a tool outcome's result weighs in its step's transcript entry: a completed one's output,
|
|
59
|
+
* a failed one's message; 0 for anything else, such as a model call's outcome.
|
|
60
|
+
*/
|
|
61
|
+
export function resultBytes(value) {
|
|
62
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
63
|
+
return 0;
|
|
64
|
+
const outcome = value;
|
|
65
|
+
if (outcome.kind === "failed")
|
|
66
|
+
return typeof outcome.message === "string" ? Buffer.byteLength(outcome.message) : 0;
|
|
67
|
+
if (outcome.kind !== "completed" || outcome.output === undefined)
|
|
68
|
+
return 0;
|
|
69
|
+
return jsonBytes(outcome.output);
|
|
70
|
+
}
|
|
71
|
+
/**
|
|
72
|
+
* False for an outcome `shapeToolOutcome` returns as it is, whatever made it: a model call's, a
|
|
73
|
+
* small tool result with no parts. Spares a look at the effect for most outcomes.
|
|
74
|
+
*/
|
|
75
|
+
export function mayNeedShaping(value) {
|
|
76
|
+
const output = value?.output;
|
|
77
|
+
if (value?.kind === "completed" && Array.isArray(output))
|
|
78
|
+
return true;
|
|
79
|
+
return resultBytes(value) > TOOL_RESULT_SMALL_BYTES;
|
|
80
|
+
}
|
|
81
|
+
/**
|
|
82
|
+
* The outcome `value` of `call`, shaped to fit its cap (see the module comment). An outcome that
|
|
83
|
+
* fits and has no binary part comes back as it is.
|
|
84
|
+
*/
|
|
85
|
+
export async function shapeToolOutcome(ctx, call, value, signal) {
|
|
86
|
+
if (!value || typeof value !== "object" || Array.isArray(value))
|
|
87
|
+
return value;
|
|
88
|
+
const outcome = value;
|
|
89
|
+
const limit = call.limit ?? TOOL_RESULT_INLINE_BYTES;
|
|
90
|
+
if (outcome.kind === "failed" && typeof outcome.message === "string")
|
|
91
|
+
return Buffer.byteLength(outcome.message) > limit
|
|
92
|
+
? { ...outcome, message: cutMessage(outcome.message, limit) }
|
|
93
|
+
: value;
|
|
94
|
+
if (outcome.kind !== "completed" || outcome.output === undefined)
|
|
95
|
+
return value;
|
|
96
|
+
const shaping = { ...call, limit, ctx, signal, files: [] };
|
|
97
|
+
let output = outcome.output;
|
|
98
|
+
if (call.mcp && isContentParts(output))
|
|
99
|
+
output = await shapeParts(shaping, output);
|
|
100
|
+
let truncated = false;
|
|
101
|
+
if (jsonBytes(output) > limit) {
|
|
102
|
+
output = await storedWhole(shaping, output);
|
|
103
|
+
truncated = true;
|
|
104
|
+
}
|
|
105
|
+
if (output === outcome.output)
|
|
106
|
+
return value;
|
|
107
|
+
return {
|
|
108
|
+
...outcome,
|
|
109
|
+
output,
|
|
110
|
+
...(shaping.files.length > 0 ? { files: shaping.files } : {}),
|
|
111
|
+
...(truncated ? { truncated: true } : {}),
|
|
112
|
+
};
|
|
113
|
+
}
|
|
114
|
+
/** True for an MCP `content` array: objects with a `type` each. */
|
|
115
|
+
function isContentParts(value) {
|
|
116
|
+
return (Array.isArray(value) &&
|
|
117
|
+
value.length > 0 &&
|
|
118
|
+
value.every((part) => !!part &&
|
|
119
|
+
typeof part === "object" &&
|
|
120
|
+
!Array.isArray(part) &&
|
|
121
|
+
typeof part.type === "string"));
|
|
122
|
+
}
|
|
123
|
+
/** An MCP result's parts: binary ones stored, then the largest text ones until the whole fits. */
|
|
124
|
+
async function shapeParts(shaping, parts) {
|
|
125
|
+
const shaped = [];
|
|
126
|
+
for (const [index, part] of parts.entries())
|
|
127
|
+
shaped.push(await shapeBinary(shaping, part, index));
|
|
128
|
+
const texts = shaped
|
|
129
|
+
.map((part, index) => ({ index, bytes: textOf(part) === undefined ? 0 : jsonBytes(part) }))
|
|
130
|
+
.filter((item) => item.bytes > 0)
|
|
131
|
+
.sort((a, b) => b.bytes - a.bytes);
|
|
132
|
+
for (const { index } of texts) {
|
|
133
|
+
if (jsonBytes(shaped) <= shaping.limit)
|
|
134
|
+
break;
|
|
135
|
+
const part = shaped[index];
|
|
136
|
+
const text = textOf(part);
|
|
137
|
+
const resource = part.type === "resource" ? part.resource : undefined;
|
|
138
|
+
const contentType = resource ? textType(resource.mimeType) : "text/plain; charset=utf-8";
|
|
139
|
+
const kept = await store(shaping, text, contentType, nameOf(shaping, contentType, index));
|
|
140
|
+
shaped[index] = withPreview({
|
|
141
|
+
type: part.type,
|
|
142
|
+
...(resource && typeof resource.uri === "string" ? { uri: resource.uri } : {}),
|
|
143
|
+
truncated: true,
|
|
144
|
+
...kept,
|
|
145
|
+
}, text, shaping.limit);
|
|
146
|
+
}
|
|
147
|
+
return shaped;
|
|
148
|
+
}
|
|
149
|
+
/** An `image`, `audio` or blob `resource` part, stored; any other part as it is. */
|
|
150
|
+
async function shapeBinary(shaping, part, index) {
|
|
151
|
+
const resource = part.type === "resource" && part.resource && typeof part.resource === "object"
|
|
152
|
+
? part.resource
|
|
153
|
+
: undefined;
|
|
154
|
+
const data = part.type === "image" || part.type === "audio"
|
|
155
|
+
? part.data
|
|
156
|
+
: resource !== undefined
|
|
157
|
+
? resource.blob
|
|
158
|
+
: undefined;
|
|
159
|
+
if (typeof data !== "string")
|
|
160
|
+
return part;
|
|
161
|
+
const declared = resource ? resource.mimeType : part.mimeType;
|
|
162
|
+
const contentType = (typeof declared === "string" ? normalizeContentType(declared) : undefined) ??
|
|
163
|
+
DEFAULT_CONTENT_TYPE;
|
|
164
|
+
const name = nameOf(shaping, contentType, index);
|
|
165
|
+
const kept = await store(shaping, Buffer.from(data, "base64"), contentType, name);
|
|
166
|
+
// An image the model reads goes to it as a file, beside the part (Q23).
|
|
167
|
+
if (part.type === "image" &&
|
|
168
|
+
"artifactId" in kept &&
|
|
169
|
+
isModelImage(contentType) &&
|
|
170
|
+
kept.size <= MODEL_FILE_MAX_BYTES)
|
|
171
|
+
shaping.files.push({
|
|
172
|
+
mediaType: contentType,
|
|
173
|
+
reference: { artifactId: kept.artifactId, version: kept.version, name },
|
|
174
|
+
});
|
|
175
|
+
return {
|
|
176
|
+
type: part.type,
|
|
177
|
+
...(resource && typeof resource.uri === "string" ? { uri: resource.uri } : {}),
|
|
178
|
+
...("artifactId" in kept ? kept : { contentType, ...kept }),
|
|
179
|
+
};
|
|
180
|
+
}
|
|
181
|
+
/** The text a part carries inline: a text part's, or a text resource's. */
|
|
182
|
+
function textOf(part) {
|
|
183
|
+
if (part.type === "text" && typeof part.text === "string")
|
|
184
|
+
return part.text;
|
|
185
|
+
if (part.type === "resource" && part.resource && typeof part.resource === "object") {
|
|
186
|
+
const text = part.resource.text;
|
|
187
|
+
if (typeof text === "string")
|
|
188
|
+
return text;
|
|
189
|
+
}
|
|
190
|
+
return undefined;
|
|
191
|
+
}
|
|
192
|
+
function textType(declared) {
|
|
193
|
+
const type = typeof declared === "string" ? normalizeContentType(declared) : undefined;
|
|
194
|
+
return type !== undefined && isText(type) ? type : "text/plain; charset=utf-8";
|
|
195
|
+
}
|
|
196
|
+
/** `value` stored whole, as text or JSON, and the preview the model gets. */
|
|
197
|
+
async function storedWhole(shaping, value) {
|
|
198
|
+
const text = typeof value === "string" ? value : JSON.stringify(value);
|
|
199
|
+
const contentType = typeof value === "string" ? "text/plain; charset=utf-8" : "application/json";
|
|
200
|
+
const kept = await store(shaping, text, contentType, nameOf(shaping, contentType));
|
|
201
|
+
return withPreview({ truncated: true, ...kept }, text, shaping.limit);
|
|
202
|
+
}
|
|
203
|
+
/**
|
|
204
|
+
* `stub` with the preview of `text`: the first 4 KiB and the last 1 KiB, or less, so that the
|
|
205
|
+
* whole fits `limit` (a step's budget spent, R2b C11); none when even a little does not.
|
|
206
|
+
*/
|
|
207
|
+
function withPreview(stub, text, limit) {
|
|
208
|
+
let budget = Math.min(PREVIEW_HEAD_BYTES + PREVIEW_TAIL_BYTES, limit);
|
|
209
|
+
for (;;) {
|
|
210
|
+
const shaped = { ...stub, preview: previewOf(text, budget) };
|
|
211
|
+
const over = jsonBytes(shaped) - limit;
|
|
212
|
+
if (over <= 0 || budget === 0)
|
|
213
|
+
return shaped;
|
|
214
|
+
budget = budget - over < 64 ? 0 : budget - over;
|
|
215
|
+
}
|
|
216
|
+
}
|
|
217
|
+
/** Stores `body` as an artifact of the call's session, or says why it was not. */
|
|
218
|
+
async function store(shaping, body, contentType, name) {
|
|
219
|
+
const bytes = typeof body === "string" ? Buffer.from(body, "utf8") : body;
|
|
220
|
+
const { ctx, request, session, signal } = shaping;
|
|
221
|
+
const ids = toolIds(request.context);
|
|
222
|
+
try {
|
|
223
|
+
const saved = await uploadArtifact(ctx, {
|
|
224
|
+
name,
|
|
225
|
+
contentType,
|
|
226
|
+
sessionId: session.id,
|
|
227
|
+
source: "engine",
|
|
228
|
+
turnId: request.turnId,
|
|
229
|
+
...("callId" in ids ? { callId: ids.callId } : {}),
|
|
230
|
+
}, bytes,
|
|
231
|
+
// Into the call's own session only.
|
|
232
|
+
{ owner: session.ownerUserId }, signal);
|
|
233
|
+
return {
|
|
234
|
+
artifactId: saved.artifact.artifactId,
|
|
235
|
+
version: saved.version.version,
|
|
236
|
+
contentType: saved.version.contentType,
|
|
237
|
+
size: saved.version.size,
|
|
238
|
+
};
|
|
239
|
+
}
|
|
240
|
+
catch (error) {
|
|
241
|
+
if (signal.aborted)
|
|
242
|
+
throw error;
|
|
243
|
+
// A limit is the Tenant's to raise; any other failure is the store's. Never the bytes.
|
|
244
|
+
ctx.config.logger.warn("tool_result_not_stored", {
|
|
245
|
+
session: session.id,
|
|
246
|
+
effect: request.effectId,
|
|
247
|
+
error: error instanceof Error ? error.message : String(error),
|
|
248
|
+
});
|
|
249
|
+
return {
|
|
250
|
+
dropped: error instanceof HttpError
|
|
251
|
+
? `The result could not be stored as an artifact: ${error.message}`
|
|
252
|
+
: "The result could not be stored as an artifact",
|
|
253
|
+
size: bytes.byteLength,
|
|
254
|
+
};
|
|
255
|
+
}
|
|
256
|
+
}
|
|
257
|
+
/** A name for a stored result: the tool's, with the part's number and the type's extension. */
|
|
258
|
+
function nameOf(shaping, contentType, index) {
|
|
259
|
+
const tool = (shaping.request.toolName ?? "tool").replace(/[^A-Za-z0-9_.-]/gu, "_");
|
|
260
|
+
const extension = extensionFor(contentType) ?? "bin";
|
|
261
|
+
return `${tool.slice(0, 200)}-result${index === undefined ? "" : `-${index + 1}`}.${extension}`;
|
|
262
|
+
}
|
|
263
|
+
/**
|
|
264
|
+
* The start and end of `text` in whole characters, `budget` bytes split four to one: by default
|
|
265
|
+
* the first `PREVIEW_HEAD_BYTES` and the last `PREVIEW_TAIL_BYTES`.
|
|
266
|
+
*/
|
|
267
|
+
export function previewOf(text, budget = PREVIEW_HEAD_BYTES + PREVIEW_TAIL_BYTES) {
|
|
268
|
+
const bytes = Buffer.from(text, "utf8");
|
|
269
|
+
if (bytes.byteLength <= budget)
|
|
270
|
+
return text;
|
|
271
|
+
if (budget <= 0)
|
|
272
|
+
return "";
|
|
273
|
+
const tailBytes = Math.floor(budget / 5);
|
|
274
|
+
const head = utf8Boundary(bytes, budget - tailBytes, "down");
|
|
275
|
+
const tail = utf8Boundary(bytes, bytes.byteLength - tailBytes, "up");
|
|
276
|
+
return `${bytes.subarray(0, head).toString("utf8")}\n…\n${bytes.subarray(tail).toString("utf8")}`;
|
|
277
|
+
}
|
|
278
|
+
/** A message past `limit`: its start and end, and how long it was. */
|
|
279
|
+
function cutMessage(message, limit) {
|
|
280
|
+
const note = `\n(The message was ${Buffer.byteLength(message)} bytes; its middle is not shown.)`;
|
|
281
|
+
const budget = Math.min(PREVIEW_HEAD_BYTES + PREVIEW_TAIL_BYTES, limit - note.length - 16);
|
|
282
|
+
return `${previewOf(message, Math.max(0, budget))}${note}`;
|
|
283
|
+
}
|
|
284
|
+
/**
|
|
285
|
+
* The nearest character boundary of the UTF-8 `bytes` at or before (`down`) or at or after
|
|
286
|
+
* (`up`) byte `at`: never inside a character.
|
|
287
|
+
*/
|
|
288
|
+
export function utf8Boundary(bytes, at, direction) {
|
|
289
|
+
let position = Math.max(0, Math.min(at, bytes.byteLength));
|
|
290
|
+
const inside = (index) => index < bytes.byteLength && (bytes[index] & 0xc0) === 0x80;
|
|
291
|
+
if (direction === "down")
|
|
292
|
+
while (position > 0 && inside(position))
|
|
293
|
+
position -= 1;
|
|
294
|
+
else
|
|
295
|
+
while (inside(position))
|
|
296
|
+
position += 1;
|
|
297
|
+
return position;
|
|
298
|
+
}
|