@halcyonlabs/app 0.47.0-halcyon.102 → 0.47.0-halcyon.110

package/dist/index.d.ts

@@ -280,8 +280,8 @@ declare const appKeybindingOverridesSchema: z$1.ZodArray<z$1.ZodObject<{
         shift: z$1.ZodBoolean;
     }, z$1.core.$strict>>;
     platform: z$1.ZodOptional<z$1.ZodEnum<{
-        mac: "mac";
         windows: "windows";
+        mac: "mac";
         linux: "linux";
     }>>;
 }, z$1.core.$strict>>;
@@ -487,8 +487,8 @@ declare const environmentSchema: z$1.ZodObject<{
         retireAt: z$1.ZodNullable<z$1.ZodNumber>;
         teardown: z$1.ZodNullable<z$1.ZodObject<{
             status: z$1.ZodEnum<{
-                running: "running";
                 failed: "failed";
+                running: "running";
                 removed: "removed";
             }>;
             attempt: z$1.ZodNumber;
@@ -583,16 +583,16 @@ declare const hostSchema: z$1.ZodObject<{
         pendingLog: z$1.ZodString;
         teardown: z$1.ZodNullable<z$1.ZodObject<{
             status: z$1.ZodEnum<{
-                running: "running";
                 failed: "failed";
+                running: "running";
                 removed: "removed";
             }>;
             attempt: z$1.ZodNumber;
         }, z$1.core.$strip>>;
     }, z$1.core.$strip>;
     maxPermissionMode: z$1.ZodEnum<{
-        auto: "auto";
         full: "full";
+        auto: "auto";
         "accept-edits": "accept-edits";
     }>;
     lastSeenAt: z$1.ZodNullable<z$1.ZodNumber>;
@@ -685,8 +685,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
     providerThreadId: z$1.ZodString;
     objective: z$1.ZodString;
     status: z$1.ZodEnum<{
-        paused: "paused";
         active: "active";
+        paused: "paused";
         budgetLimited: "budgetLimited";
         complete: "complete";
     }>;
@@ -1257,9 +1257,9 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
             step: z$1.ZodString;
             status: z$1.ZodOptional<z$1.ZodEnum<{
                 pending: "pending";
+                active: "active";
                 completed: "completed";
                 failed: "failed";
-                active: "active";
             }>>;
         }, z$1.core.$strip>>;
         explanation: z$1.ZodOptional<z$1.ZodString>;
@@ -1338,10 +1338,10 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         }>;
         taskStatus: z$1.ZodEnum<{
             pending: "pending";
-            running: "running";
-            paused: "paused";
             completed: "completed";
             failed: "failed";
+            paused: "paused";
+            running: "running";
             killed: "killed";
             stopped: "stopped";
         }>;
@@ -1358,8 +1358,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 index: z$1.ZodNumber;
                 label: z$1.ZodString;
                 state: z$1.ZodEnum<{
-                    running: "running";
                     failed: "failed";
+                    running: "running";
                     queued: "queued";
                     done: "done";
                     skipped: "skipped";
@@ -2029,9 +2029,9 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
             step: z$1.ZodString;
             status: z$1.ZodOptional<z$1.ZodEnum<{
                 pending: "pending";
+                active: "active";
                 completed: "completed";
                 failed: "failed";
-                active: "active";
             }>>;
         }, z$1.core.$strip>>;
         explanation: z$1.ZodOptional<z$1.ZodString>;
@@ -2110,10 +2110,10 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         }>;
         taskStatus: z$1.ZodEnum<{
             pending: "pending";
-            running: "running";
-            paused: "paused";
             completed: "completed";
             failed: "failed";
+            paused: "paused";
+            running: "running";
             killed: "killed";
             stopped: "stopped";
         }>;
@@ -2130,8 +2130,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 index: z$1.ZodNumber;
                 label: z$1.ZodString;
                 state: z$1.ZodEnum<{
-                    running: "running";
                     failed: "failed";
+                    running: "running";
                     queued: "queued";
                     done: "done";
                     skipped: "skipped";
@@ -2342,10 +2342,10 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         }>;
         taskStatus: z$1.ZodEnum<{
             pending: "pending";
-            running: "running";
-            paused: "paused";
             completed: "completed";
             failed: "failed";
+            paused: "paused";
+            running: "running";
             killed: "killed";
             stopped: "stopped";
         }>;
@@ -2362,8 +2362,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 index: z$1.ZodNumber;
                 label: z$1.ZodString;
                 state: z$1.ZodEnum<{
-                    running: "running";
                     failed: "failed";
+                    running: "running";
                     queued: "queued";
                     done: "done";
                     skipped: "skipped";
@@ -2441,10 +2441,10 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         }>;
         taskStatus: z$1.ZodEnum<{
             pending: "pending";
-            running: "running";
-            paused: "paused";
             completed: "completed";
             failed: "failed";
+            paused: "paused";
+            running: "running";
             killed: "killed";
             stopped: "stopped";
         }>;
@@ -2461,8 +2461,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 index: z$1.ZodNumber;
                 label: z$1.ZodString;
                 state: z$1.ZodEnum<{
-                    running: "running";
                     failed: "failed";
+                    running: "running";
                     queued: "queued";
                     done: "done";
                     skipped: "skipped";
@@ -2657,9 +2657,9 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         step: z$1.ZodString;
         status: z$1.ZodOptional<z$1.ZodEnum<{
             pending: "pending";
+            active: "active";
             completed: "completed";
             failed: "failed";
-            active: "active";
         }>>;
     }, z$1.core.$strip>>;
     explanation: z$1.ZodOptional<z$1.ZodString>;
@@ -2668,8 +2668,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
     threadId: z$1.ZodString;
     providerThreadId: z$1.ZodString;
     state: z$1.ZodEnum<{
-        running: "running";
         idle: "idle";
+        running: "running";
     }>;
 }, z$1.core.$strip>, z$1.ZodObject<{
     type: z$1.ZodLiteral<"turn/diff/updated">;
@@ -2961,12 +2961,12 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 }>;
                 name: z$1.ZodString;
                 source: z$1.ZodEnum<{
-                    command: "command";
                     skill: "skill";
+                    command: "command";
                 }>;
                 origin: z$1.ZodEnum<{
-                    project: "project";
                     user: "user";
+                    project: "project";
                     builtin: "builtin";
                 }>;
                 label: z$1.ZodString;
@@ -3051,12 +3051,12 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
                 }>;
                 name: z$1.ZodString;
                 source: z$1.ZodEnum<{
-                    command: "command";
                     skill: "skill";
+                    command: "command";
                 }>;
                 origin: z$1.ZodEnum<{
-                    project: "project";
                     user: "user";
+                    project: "project";
                     builtin: "builtin";
                 }>;
                 label: z$1.ZodString;
@@ -3129,8 +3129,8 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
         }>;
         permissionMode: z$1.ZodEnum<{
             readonly: "readonly";
-            auto: "auto";
             full: "full";
+            auto: "auto";
             "accept-edits": "accept-edits";
             "workspace-write": "workspace-write";
         }>;
@@ -3531,9 +3531,9 @@ declare const threadEventSchema: z$1.ZodUnion<readonly [z$1.ZodIntersection<z$1.
     threadId: z$1.ZodString;
     provisioningId: z$1.ZodString;
     status: z$1.ZodEnum<{
+        active: "active";
         completed: "completed";
         failed: "failed";
-        active: "active";
         cancelled: "cancelled";
     }>;
     environmentId: z$1.ZodNullable<z$1.ZodString>;
@@ -3928,8 +3928,8 @@ type ReasoningLevel = z$1.infer<typeof reasoningLevelSchema>;
 declare const serviceTierSchema: z$1.ZodString;
 type ServiceTier = z$1.infer<typeof serviceTierSchema>;
 declare const permissionModeSchema: z$1.ZodEnum<{
-    auto: "auto";
     full: "full";
+    auto: "auto";
     "accept-edits": "accept-edits";
 }>;
 type PermissionMode = z$1.infer<typeof permissionModeSchema>;
@@ -3969,12 +3969,12 @@ declare const promptTextMentionSchema: z$1.ZodObject<{
         }>;
         name: z$1.ZodString;
         source: z$1.ZodEnum<{
-            command: "command";
             skill: "skill";
+            command: "command";
         }>;
         origin: z$1.ZodEnum<{
-            project: "project";
             user: "user";
+            project: "project";
             builtin: "builtin";
         }>;
         label: z$1.ZodString;
@@ -4034,12 +4034,12 @@ declare const promptInputSchema: z$1.ZodDiscriminatedUnion<[z$1.ZodObject<{
             }>;
             name: z$1.ZodString;
             source: z$1.ZodEnum<{
-                command: "command";
                 skill: "skill";
+                command: "command";
             }>;
             origin: z$1.ZodEnum<{
-                project: "project";
                 user: "user";
+                project: "project";
                 builtin: "builtin";
             }>;
             label: z$1.ZodString;
@@ -4098,8 +4098,8 @@ declare const resolvedThreadExecutionOptionsSchema: z$1.ZodObject<{
     serviceTier: z$1.ZodString;
     reasoningLevel: z$1.ZodString;
     permissionMode: z$1.ZodEnum<{
-        auto: "auto";
         full: "full";
+        auto: "auto";
         "accept-edits": "accept-edits";
     }>;
     source: z$1.ZodEnum<{
@@ -4115,8 +4115,8 @@ declare const projectExecutionDefaultsSchema: z$1.ZodObject<{
     serviceTier: z$1.ZodString;
     reasoningLevel: z$1.ZodString;
     permissionMode: z$1.ZodEnum<{
-        auto: "auto";
         full: "full";
+        auto: "auto";
         "accept-edits": "accept-edits";
     }>;
 }, z$1.core.$strip>;
@@ -4156,8 +4156,8 @@ declare const providerInfoSchema: z$1.ZodObject<{
         supportsFork: z$1.ZodBoolean;
         supportsSessionRewind: z$1.ZodBoolean;
         permissionModes: z$1.ZodArray<z$1.ZodEnum<{
-            auto: "auto";
             full: "full";
+            auto: "auto";
             "accept-edits": "accept-edits";
         }>>;
         modelCatalogScope: z$1.ZodEnum<{
@@ -4346,12 +4346,12 @@ declare const threadQueuedMessageSchema: z$1.ZodObject<{
                 }>;
                 name: z$1.ZodString;
                 source: z$1.ZodEnum<{
-                    command: "command";
                     skill: "skill";
+                    command: "command";
                 }>;
                 origin: z$1.ZodEnum<{
-                    project: "project";
                     user: "user";
+                    project: "project";
                     builtin: "builtin";
                 }>;
                 label: z$1.ZodString;
@@ -4397,8 +4397,8 @@ declare const threadQueuedMessageSchema: z$1.ZodObject<{
     model: z$1.ZodString;
     reasoningLevel: z$1.ZodString;
     permissionMode: z$1.ZodEnum<{
-        auto: "auto";
         full: "full";
+        auto: "auto";
         "accept-edits": "accept-edits";
     }>;
     serviceTier: z$1.ZodString;
@@ -7456,8 +7456,8 @@ declare const hostDaemonCommandRegistry: {
             skillFilePath: z$1.ZodString;
         }, z$1.core.$strict>], "kind">>;
         instructionMode: z$1.ZodEnum<{
-            append: "append";
             replace: "replace";
+            append: "append";
         }>;
         type: z$1.ZodLiteral<"thread.rewind.prepare">;
         leaseId: z$1.ZodString;
@@ -7605,8 +7605,8 @@ declare const hostDaemonCommandRegistry: {
             skillFilePath: z$1.ZodString;
         }, z$1.core.$strict>], "kind">>;
         instructionMode: z$1.ZodEnum<{
-            append: "append";
             replace: "replace";
+            append: "append";
         }>;
         type: z$1.ZodLiteral<"thread.start">;
         requestId: z$1.ZodString;
@@ -8069,8 +8069,8 @@ declare const hostDaemonCommandRegistry: {
                 envPassthrough: z$1.ZodArray<z$1.ZodString>;
             }, z$1.core.$strict>;
             instructionMode: z$1.ZodEnum<{
-                append: "append";
                 replace: "replace";
+                append: "append";
             }>;
             projectId: z$1.ZodString;
             providerId: z$1.ZodString;
@@ -8285,8 +8285,8 @@ declare const hostDaemonCommandRegistry: {
                 envPassthrough: z$1.ZodArray<z$1.ZodString>;
             }, z$1.core.$strict>;
             instructionMode: z$1.ZodEnum<{
-                append: "append";
                 replace: "replace";
+                append: "append";
             }>;
             projectId: z$1.ZodString;
             providerId: z$1.ZodString;
@@ -9187,13 +9187,13 @@ declare const hostDaemonCommandRegistry: {
         modifiedAtMs: z$1.ZodOptional<z$1.ZodNumber>;
         sha256: z$1.ZodString;
     }, z$1.core.$strip>, z$1.ZodObject<{
+        path: z$1.ZodString;
         contentEncoding: z$1.ZodEnum<{
             base64: "base64";
             utf8: "utf8";
         }>;
         sha256: z$1.ZodString;
         sizeBytes: z$1.ZodNumber;
-        path: z$1.ZodString;
         mimeType: z$1.ZodOptional<z$1.ZodString>;
         modifiedAtMs: z$1.ZodOptional<z$1.ZodNumber>;
         notModified: z$1.ZodLiteral<true>;
@@ -9511,8 +9511,8 @@ declare const hostDaemonCommandRegistry: {
         installed: z$1.ZodBoolean;
         installSource: z$1.ZodEnum<{
             notInstalled: "notInstalled";
-            external: "external";
             npmGlobal: "npmGlobal";
+            external: "external";
         }>;
         currentVersion: z$1.ZodNullable<z$1.ZodString>;
         latestVersion: z$1.ZodNullable<z$1.ZodString>;
@@ -10062,8 +10062,8 @@ declare const providerCliStatusResponseSchema: z$1.ZodRecord<z$1.ZodString, z$1.
     installed: z$1.ZodBoolean;
     installSource: z$1.ZodEnum<{
         notInstalled: "notInstalled";
-        external: "external";
         npmGlobal: "npmGlobal";
+        external: "external";
     }>;
     currentVersion: z$1.ZodNullable<z$1.ZodString>;
     latestVersion: z$1.ZodNullable<z$1.ZodString>;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@halcyonlabs/app",
-  "version": "0.47.0-halcyon.102",
+  "version": "0.47.0-halcyon.110",
   "description": "bb app launcher for npx",
   "keywords": [
     "bb",

package/server/dist/builtin-plugins/orbit/dist/assets/orbit/bundle/scripts/orbit-job-supervisor

@@ -14,8 +14,7 @@ import re
 # A job has no size or file-count cap of its own. Work stops only when the disk
 # holding job folders falls below MAIN_FREE_FLOOR, which keeps room for the OS,
 # bb and Orbit's own state. A new job starts only with START_HEADROOM above it.
-RETAINED_JOB_BYTES = 1_000_000_000
-RETAINED_TOTAL_BYTES = 8_000_000_000
+RESULT_SPACE_FLOOR = 1_000_000
 MAIN_FREE_FLOOR = 15_000_000_000
 START_HEADROOM = 5_000_000_000
 # Many tiny files can exhaust inodes while bytes remain: the same rule applies
@@ -27,11 +26,11 @@ MAX_CONCURRENT_JOBS = 8
 JOB_MEMORY_BYTES = 6 * 2**30
 DESKTOP_MEMORY_BYTES = 6 * 2**30
 CAPACITY_FULL_STATUS = 75
-LIMITS = dict(retained_bytes=RETAINED_JOB_BYTES)
+LIMITS = {}
 # Records written before the free-space rule carry the old fixed caps. They load
 # as current records and are rewritten in that form at their next transition.
-LEGACY_LIMITS = dict(bytes=6_000_000_000, files=250_000, retained_bytes=RETAINED_JOB_BYTES)
-assert START_HEADROOM >= RETAINED_JOB_BYTES
+LEGACY_LIMITS = dict(bytes=6_000_000_000, files=250_000, retained_bytes=1_000_000_000)
+PREVIOUS_LIMITS = dict(retained_bytes=1_000_000_000)
 STATES = ('reserved', 'running', 'parked', 'stopping', 'exporting', 'destroying', 'closed')
 ENDS = ('finished', 'failed', 'cancelled', 'timed_out', 'crashed', 'storage_limit')
 
@@ -104,6 +103,10 @@ def new_manifest(job_id, owner, outputs, *, workspace=True):
     return validate_manifest(manifest)
 
 
+def result_metadata_space(manifest):
+    return RESULT_SPACE_FLOOR + sum(len(name.encode('utf-8')) * 6 + 512 for name in manifest['outputs'])
+
+
 def has_workspace(manifest):
     return manifest.get('workspace', True)
 
@@ -120,7 +123,7 @@ def validate_manifest(manifest):
             or not isinstance(manifest['owner'], str)
             or re.fullmatch('thr_[A-Za-z0-9_-]{1,128}', manifest['owner']) is None
             or manifest['state'] not in STATES
-            or manifest['limits'] not in (LIMITS, LEGACY_LIMITS)):
+            or manifest['limits'] not in (LIMITS, LEGACY_LIMITS, PREVIOUS_LIMITS)):
         raise ValueError('Invalid job record.')
     ending = manifest['state'] in ('stopping', 'exporting', 'destroying', 'closed')
     if (ending and manifest['end'] not in ENDS) or (not ending and manifest['end'] is not None):
@@ -324,7 +327,7 @@ class Workspaces:
         # or thread state. A corrupt or unreadable entry blocks admission.
         return [self.read(name[:-5]) for name in os.listdir(self.state) if name.endswith('.json')]
 
-    def create(self, owner, outputs, *, free, results_free=None, inodes=None, job_id=None, workspace=True):
+    def create(self, owner, outputs, *, free=None, results_free=None, inodes=None, job_id=None, workspace=True):
         manifest = new_manifest(job_id if job_id is not None else secrets.token_hex(16), owner, outputs, workspace=workspace)
         with self.locked():
             records = self.records()
@@ -339,14 +342,17 @@ class Workspaces:
             slots = [r for r in records if holds_slot(r)]
             if len(slots) >= concurrent_jobs(self.state):
                 raise CapacityFull('Job capacity is reserved; wait for disposal.')
+            if free is None:
+                free = work_free(self)
+                if workspace:
+                    results_free = available(self.results)
+                    inodes = work_inodes(self)
             if not workspace:
                 if type(free) is not int or free < MAIN_FREE_FLOOR:
                     raise CapacityFull(f'Station disk has {gigabytes(free)} GB free space; a start needs '
                                        f'{gigabytes(MAIN_FREE_FLOOR)} GB. Archive handled work to free space.')
             else:
-                retained = sum(self.result_reservation(r) for r in records)
-                if retained + RETAINED_JOB_BYTES > RETAINED_TOTAL_BYTES:
-                    raise CapacityFull('Result capacity is reserved.')
+                retained = sum(self.result_reservation(r) for r in records) + result_metadata_space(manifest)
                 if type(free) is not int or free < MAIN_FREE_FLOOR + START_HEADROOM:
                     raise CapacityFull(f'Station disk has {gigabytes(free)} GB free space; a new job needs '
                                        f'{gigabytes(MAIN_FREE_FLOOR + START_HEADROOM)} GB. Archive handled work to free space.')
@@ -354,9 +360,9 @@ class Workspaces:
                     raise CapacityFull(f'Station disk has {inodes} free file entries; a new job needs '
                                        f'{INODE_FLOOR + START_INODE_HEADROOM}. Archive handled work to free space.')
                 results_free = free if results_free is None else results_free
-                if type(results_free) is not int or results_free < MAIN_FREE_FLOOR + RETAINED_JOB_BYTES:
+                if type(results_free) is not int or results_free < MAIN_FREE_FLOOR + retained:
                     raise CapacityFull(f'The saved-results disk has {gigabytes(results_free)} GB free space; a new job needs '
-                                       f'{gigabytes(MAIN_FREE_FLOOR + RETAINED_JOB_BYTES)} GB there. Free space on it first.')
+                                       f'{gigabytes(MAIN_FREE_FLOOR + retained)} GB there. Free space on it first.')
             # Persist ownership first. A crash before mkdir still leaves a reserved
             # attempt which restart must explicitly finish, never silently forget.
             write_record(self.state, manifest['id'] + '.json', manifest, exclusive=True)
@@ -519,19 +525,39 @@ class Workspaces:
                     low=low, bytes=size, files=files, complete=complete)
 
     def result_reservation(self, manifest):
-        if not has_workspace(manifest):
+        if not has_workspace(manifest) or not holds_capacity(manifest):
             return 0
-        if holds_capacity(manifest):
-            return RETAINED_JOB_BYTES
-        fd = os.open(manifest['id'] + '.retained', os.O_RDONLY | os.O_NOFOLLOW | os.O_CLOEXEC, dir_fd=self.state)
-        with os.fdopen(fd, 'rb') as stream:
-            metadata = os.fstat(stream.fileno())
-            if not stat.S_ISREG(metadata.st_mode) or metadata.st_size > 1024 or metadata.st_nlink != 1:
-                raise ValueError('Invalid retained-size record.')
-            saved = json.load(stream)
-        if set(saved) != {'bytes'} or type(saved['bytes']) is not int or not 0 <= saved['bytes'] <= RETAINED_JOB_BYTES:
-            raise ValueError('Invalid retained-size record.')
-        return saved['bytes']
+        manifest = self.declared_outputs(manifest, persist=False)
+        total = result_metadata_space(manifest)
+        try:
+            root = self.open_workspace(manifest['id'])
+        except FileNotFoundError:
+            return total
+        try:
+            for name in manifest['outputs']:
+                try:
+                    fd = open_output(root, name)
+                except UnsafeOutput:
+                    raise
+                except (FileNotFoundError, ValueError):
+                    continue
+                except OSError as error:
+                    if error.errno not in (errno.ELOOP, errno.ENOTDIR):
+                        raise
+                    continue
+                try:
+                    total += os.fstat(fd).st_size
+                finally:
+                    os.close(fd)
+        finally:
+            os.close(root)
+        return total
+
+    def require_result_space(self, required):
+        free = available(self.results)
+        if free < MAIN_FREE_FLOOR + required:
+            raise CapacityFull(f'The saved-results disk has {gigabytes(free)} GB free space; saving results needs '
+                               f'{gigabytes(MAIN_FREE_FLOOR + required)} GB there. Free space on it first.')
 
     def read_output_declaration(self, root):
         fd = open_output(root, '.orbit-outputs.json')
@@ -546,7 +572,7 @@ class Workspaces:
         finally:
             os.close(fd)
 
-    def declared_outputs(self, manifest):
+    def declared_outputs(self, manifest, *, persist=True):
         try:
             root = self.open_workspace(manifest['id'])
         except FileNotFoundError:
@@ -565,7 +591,8 @@ class Workspaces:
         if warned:
             updated['outputWarning'] = OUTPUT_WARNING
         updated = validate_manifest(updated)
-        write_record(self.state, manifest['id'] + '.json', updated)
+        if persist:
+            write_record(self.state, manifest['id'] + '.json', updated)
         return updated
 
     def park(self, job_id):
@@ -677,8 +704,8 @@ class Workspaces:
             if 'outputs.json' in os.listdir(destination):
                 self.verify_outputs(manifest)
                 return
+            self.require_result_space(self.result_reservation(manifest))
             receipts = []
-            total = 0
             for index, name in enumerate(manifest['outputs']):
                 receipt = dict(name=name, status='missing')
                 if source is None:
@@ -703,8 +730,7 @@ class Workspaces:
                     continue
                 try:
                     metadata = os.fstat(fd)
-                    if total + metadata.st_size > RETAINED_JOB_BYTES:
-                        raise ValueError('Declared results exceed their allowance; preserve and increase the allowance.')
+                    self.require_result_space(metadata.st_size + RESULT_SPACE_FLOOR)
                     number = format(index, '04d')
                     temporary = '.' + number
                     # Private partial exports can be replaced on restart. The
@@ -721,8 +747,7 @@ class Workspaces:
                             if not block:
                                 break
                             count += len(block)
-                            if total + count > RETAINED_JOB_BYTES:
-                                raise ValueError('Declared results exceed their allowance.')
+                            self.require_result_space(len(block) + RESULT_SPACE_FLOOR)
                             digest.update(block)
                             stream.write(block)
                         if file_identity(os.fstat(fd)) != file_identity(metadata):
@@ -731,7 +756,6 @@ class Workspaces:
                         os.fchmod(stream.fileno(), 0o444)
                         os.fsync(stream.fileno())
                     os.rename(temporary, number, src_dir_fd=destination, dst_dir_fd=destination)
-                    total += count
                     receipt.update(status='saved', file=number, bytes=count, sha256=digest.hexdigest())
                     receipts.append(receipt)
                 finally:
@@ -778,12 +802,8 @@ class Workspaces:
                         if not block:
                             break
                         count += len(block)
-                        if count > RETAINED_JOB_BYTES:
-                            raise ValueError('Retained result exceeds its allowance.')
                         digest.update(block)
                     total += count
-                    if total > RETAINED_JOB_BYTES:
-                        raise ValueError('Retained results exceed their allowance.')
                     if count != item['bytes'] or digest.hexdigest() != item['sha256']:
                         raise ValueError('Retained result bytes changed; workspace preserved.')
                 finally:
@@ -1304,12 +1324,10 @@ def managed_request(request):
 
 def managed_operation(store, account, request):
     if request['operation'] == 'reserve':
-        store.create(request['owner'], [], free=work_free(store), job_id=request['id'], workspace=False)
+        store.create(request['owner'], [], job_id=request['id'], workspace=False)
     elif request['operation'] == 'create':
         declared_output_names(request['outputs'], strict=True)
-        job = store.create(request['owner'], request['outputs'], free=work_free(store),
-                           results_free=available(store.results), inodes=work_inodes(store),
-                           job_id=request['id'])
+        job = store.create(request['owner'], request['outputs'], job_id=request['id'])
         root = store.open_workspace(job['id'])
         try:
             prepare_hooks(root)

package/server/dist/builtin-plugins/orbit/dist/assets/orbit/recipes/base-linux/antfly.sh

@@ -57,9 +57,9 @@ After=network.target
 User=vectors
 Group=vectors
 Environment=HOME=/var/db/orbit/vectors
-# The embedding model is loaded at start (about two minutes on two vCPUs): loaded on
-# demand, the first index creation's probe times out while it loads.
-ExecStart=/usr/local/bin/antfly standalone --host 127.0.0.1 --port 8080 --health-port 4200 --data-dir $data --models-dir $models --preload-model embedder:$model
+ExecStart=/usr/local/bin/antfly standalone --host 127.0.0.1 --port 8080 --health-port 4200 --data-dir $data --models-dir $models --process-memory-budget-mb 6144 --inference-kv-budget-mb 64 --preload-model embedder:$model
+MemoryAccounting=true
+MemoryMax=6144M
 Restart=always
 RestartSec=3
 NoNewPrivileges=true