omnigateway 0.1.3 → 0.1.4

package/bin/omni.js

@@ -17335,6 +17335,27 @@ ALTER TABLE credentials ADD COLUMN disabled_reason TEXT;
 ALTER TABLE credentials ADD COLUMN disabled_at INTEGER;
 `;
 
+// packages/store/src/sqlite/migrations/004_request_state.sql
+var _004_request_state_default = `-- Whether a request is still running.
+--
+-- A row used to be written once, after the response stream drained, so a long
+-- stream was invisible for its whole life and the console showed an idle
+-- gateway. A row is now written twice: \`pending\` when dispatch starts, \`done\`
+-- when it finishes.
+--
+-- Existing rows default to 'done'. Every row that predates this migration
+-- describes a request that has already ended, by definition.
+ALTER TABLE request_logs ADD COLUMN state TEXT NOT NULL DEFAULT 'done';
+
+-- \`status\` and \`duration_ms\` stay NOT NULL: making them nullable means a table
+-- rebuild in SQLite, and a pending row's zeros are never read. Readers key off
+-- \`state\` alone.
+--
+-- Partial, because pending rows are a handful at any moment while the table
+-- holds a month of finished ones.
+CREATE INDEX idx_request_logs_pending ON request_logs(state) WHERE state = 'pending';
+`;
+
 // packages/store/src/sqlite/rollup.ts
 function startOfLocalDay(at) {
   const day = new Date(at);
@@ -17441,7 +17462,8 @@ function backfillDaily(db) {
 var MIGRATIONS = [
   { id: 1, sql: _001_init_default },
   { id: 2, sql: _002_usage_daily_default, after: backfillDaily },
-  { id: 3, sql: _003_quota_snapshot_default }
+  { id: 3, sql: _003_quota_snapshot_default },
+  { id: 4, sql: _004_request_state_default }
 ];
 function openDb(path) {
   const db = new Database(path, { create: true });
@@ -17513,6 +17535,7 @@ function createKeyRepo(db) {
 // packages/store/src/sqlite/usage.ts
 var toLog = (r) => ({
   id: r.id,
+  state: r.state === "pending" ? "pending" : "done",
   at: r.at,
   apiKeyId: r.api_key_id,
   requestedModel: r.requested_model,
@@ -17571,37 +17594,78 @@ function label(value) {
   const text = String(value);
   return text.length === 0 ? "unknown" : text;
 }
+var COLUMNS = `(id, state, at, api_key_id, requested_model, resolved_provider, resolved_model,
+                  credential_id, attempts, status, error_code, input_tokens, output_tokens,
+                  cache_read_tokens, cache_write_tokens, ttft_ms, duration_ms, cost_usd,
+                  degradations)`;
+var PLACEHOLDERS = "(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)";
+function values(log, state) {
+  return [
+    log.id,
+    state,
+    log.at,
+    log.apiKeyId,
+    log.requestedModel,
+    log.resolvedProvider,
+    log.resolvedModel,
+    log.credentialId,
+    log.attempts,
+    log.status,
+    log.errorCode,
+    log.inputTokens,
+    log.outputTokens,
+    log.cacheReadTokens,
+    log.cacheWriteTokens,
+    log.ttftMs,
+    log.durationMs,
+    log.costUsd,
+    JSON.stringify(log.degradations)
+  ];
+}
+var COMPLETE = `INSERT INTO request_logs ${COLUMNS} VALUES ${PLACEHOLDERS}
+  ON CONFLICT(id) DO UPDATE SET
+    state = 'done',
+    requested_model = COALESCE(NULLIF(excluded.requested_model, ''), request_logs.requested_model),
+    api_key_id = COALESCE(excluded.api_key_id, request_logs.api_key_id),
+    resolved_provider = excluded.resolved_provider,
+    resolved_model = excluded.resolved_model,
+    credential_id = excluded.credential_id,
+    attempts = excluded.attempts,
+    status = excluded.status,
+    error_code = excluded.error_code,
+    input_tokens = excluded.input_tokens,
+    output_tokens = excluded.output_tokens,
+    cache_read_tokens = excluded.cache_read_tokens,
+    cache_write_tokens = excluded.cache_write_tokens,
+    ttft_ms = excluded.ttft_ms,
+    duration_ms = excluded.duration_ms,
+    cost_usd = excluded.cost_usd,
+    degradations = excluded.degradations`;
 function createUsageRepo(db) {
-  const insert = db.transaction((log) => {
-    db.run(`INSERT INTO request_logs
-           (id, at, api_key_id, requested_model, resolved_provider, resolved_model, credential_id,
-            attempts, status, error_code, input_tokens, output_tokens, cache_read_tokens,
-            cache_write_tokens, ttft_ms, duration_ms, cost_usd, degradations)
-         VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)`, [
-      log.id,
-      log.at,
-      log.apiKeyId,
-      log.requestedModel,
-      log.resolvedProvider,
-      log.resolvedModel,
-      log.credentialId,
-      log.attempts,
-      log.status,
-      log.errorCode,
-      log.inputTokens,
-      log.outputTokens,
-      log.cacheReadTokens,
-      log.cacheWriteTokens,
-      log.ttftMs,
-      log.durationMs,
-      log.costUsd,
-      JSON.stringify(log.degradations)
-    ]);
-    rollupLog(db, log);
+  const complete = db.transaction((log) => {
+    db.run(COMPLETE, values(log, "done"));
+    const stored = db.query("SELECT * FROM request_logs WHERE id = ?").get(log.id);
+    if (stored !== null)
+      rollupLog(db, toLog(stored));
   });
   return {
+    async begin(log) {
+      db.run(`INSERT INTO request_logs ${COLUMNS} VALUES ${PLACEHOLDERS}`, values(log, "pending"));
+    },
+    async route(id, target) {
+      db.run(`UPDATE request_logs
+            SET resolved_provider = ?, resolved_model = ?, credential_id = ?
+          WHERE id = ? AND state = 'pending'`, [target.provider, target.model, target.credentialId, id]);
+    },
     async append(log) {
-      insert(log);
+      complete(log);
+    },
+    async sweepPending() {
+      const stale = db.query("SELECT * FROM request_logs WHERE state = 'pending'").all().map(toLog);
+      for (const log of stale) {
+        complete({ ...log, status: 499, errorCode: "interrupted", durationMs: 0 });
+      }
+      return stale.length;
     },
     async recent(limit) {
       return db.query("SELECT * FROM request_logs ORDER BY at DESC LIMIT ?").all(limit).map(toLog);
@@ -17623,7 +17687,7 @@ function createUsageRepo(db) {
                   COALESCE(SUM(cache_write_tokens), 0) AS cache_write_tokens,
                   COALESCE(SUM(cost_usd), 0) AS cost_usd
              FROM ${daily ? "usage_daily" : "request_logs"}
-            WHERE ${timeColumn} >= ? AND ${timeColumn} <= ?
+            WHERE ${daily ? "" : "state = 'done' AND "}${timeColumn} >= ? AND ${timeColumn} <= ?
             GROUP BY key${split === null ? "" : ", split"}
             ORDER BY requests DESC`).all(since, until);
       return rows.map((r) => {
@@ -18102,21 +18166,21 @@ function parse5(argv, options = {}) {
     throw new UsageError(error51 instanceof Error ? error51.message : "could not parse arguments");
   }
 }
-function stringFlag(values, name) {
-  const value = values[name];
+function stringFlag(values2, name) {
+  const value = values2[name];
   return typeof value === "string" ? value : undefined;
 }
-function boolFlag(values, name) {
-  return values[name] === true;
+function boolFlag(values2, name) {
+  return values2[name] === true;
 }
-function listFlag(values, name) {
-  const value = values[name];
+function listFlag(values2, name) {
+  const value = values2[name];
   if (Array.isArray(value))
     return value.filter((entry) => typeof entry === "string");
   return typeof value === "string" ? [value] : undefined;
 }
-function numberFlag(values, name) {
-  const raw = stringFlag(values, name);
+function numberFlag(values2, name) {
+  const raw = stringFlag(values2, name);
   if (raw === undefined)
     return;
   const value = Number(raw);

package/gateway.js

@@ -22634,6 +22634,27 @@ ALTER TABLE credentials ADD COLUMN disabled_reason TEXT;
 ALTER TABLE credentials ADD COLUMN disabled_at INTEGER;
 `;
 
+// packages/store/src/sqlite/migrations/004_request_state.sql
+var _004_request_state_default = `-- Whether a request is still running.
+--
+-- A row used to be written once, after the response stream drained, so a long
+-- stream was invisible for its whole life and the console showed an idle
+-- gateway. A row is now written twice: \`pending\` when dispatch starts, \`done\`
+-- when it finishes.
+--
+-- Existing rows default to 'done'. Every row that predates this migration
+-- describes a request that has already ended, by definition.
+ALTER TABLE request_logs ADD COLUMN state TEXT NOT NULL DEFAULT 'done';
+
+-- \`status\` and \`duration_ms\` stay NOT NULL: making them nullable means a table
+-- rebuild in SQLite, and a pending row's zeros are never read. Readers key off
+-- \`state\` alone.
+--
+-- Partial, because pending rows are a handful at any moment while the table
+-- holds a month of finished ones.
+CREATE INDEX idx_request_logs_pending ON request_logs(state) WHERE state = 'pending';
+`;
+
 // packages/store/src/sqlite/rollup.ts
 function startOfLocalDay(at) {
   const day = new Date(at);
@@ -22740,7 +22761,8 @@ function backfillDaily(db) {
 var MIGRATIONS = [
   { id: 1, sql: _001_init_default },
   { id: 2, sql: _002_usage_daily_default, after: backfillDaily },
-  { id: 3, sql: _003_quota_snapshot_default }
+  { id: 3, sql: _003_quota_snapshot_default },
+  { id: 4, sql: _004_request_state_default }
 ];
 function openDb(path) {
   const db = new Database(path, { create: true });
@@ -22812,6 +22834,7 @@ function createKeyRepo(db) {
 // packages/store/src/sqlite/usage.ts
 var toLog = (r) => ({
   id: r.id,
+  state: r.state === "pending" ? "pending" : "done",
   at: r.at,
   apiKeyId: r.api_key_id,
   requestedModel: r.requested_model,
@@ -22870,37 +22893,78 @@ function label(value) {
   const text = String(value);
   return text.length === 0 ? "unknown" : text;
 }
+var COLUMNS = `(id, state, at, api_key_id, requested_model, resolved_provider, resolved_model,
+                  credential_id, attempts, status, error_code, input_tokens, output_tokens,
+                  cache_read_tokens, cache_write_tokens, ttft_ms, duration_ms, cost_usd,
+                  degradations)`;
+var PLACEHOLDERS = "(?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)";
+function values(log, state) {
+  return [
+    log.id,
+    state,
+    log.at,
+    log.apiKeyId,
+    log.requestedModel,
+    log.resolvedProvider,
+    log.resolvedModel,
+    log.credentialId,
+    log.attempts,
+    log.status,
+    log.errorCode,
+    log.inputTokens,
+    log.outputTokens,
+    log.cacheReadTokens,
+    log.cacheWriteTokens,
+    log.ttftMs,
+    log.durationMs,
+    log.costUsd,
+    JSON.stringify(log.degradations)
+  ];
+}
+var COMPLETE = `INSERT INTO request_logs ${COLUMNS} VALUES ${PLACEHOLDERS}
+  ON CONFLICT(id) DO UPDATE SET
+    state = 'done',
+    requested_model = COALESCE(NULLIF(excluded.requested_model, ''), request_logs.requested_model),
+    api_key_id = COALESCE(excluded.api_key_id, request_logs.api_key_id),
+    resolved_provider = excluded.resolved_provider,
+    resolved_model = excluded.resolved_model,
+    credential_id = excluded.credential_id,
+    attempts = excluded.attempts,
+    status = excluded.status,
+    error_code = excluded.error_code,
+    input_tokens = excluded.input_tokens,
+    output_tokens = excluded.output_tokens,
+    cache_read_tokens = excluded.cache_read_tokens,
+    cache_write_tokens = excluded.cache_write_tokens,
+    ttft_ms = excluded.ttft_ms,
+    duration_ms = excluded.duration_ms,
+    cost_usd = excluded.cost_usd,
+    degradations = excluded.degradations`;
 function createUsageRepo(db) {
-  const insert = db.transaction((log) => {
-    db.run(`INSERT INTO request_logs
-           (id, at, api_key_id, requested_model, resolved_provider, resolved_model, credential_id,
-            attempts, status, error_code, input_tokens, output_tokens, cache_read_tokens,
-            cache_write_tokens, ttft_ms, duration_ms, cost_usd, degradations)
-         VALUES (?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?,?)`, [
-      log.id,
-      log.at,
-      log.apiKeyId,
-      log.requestedModel,
-      log.resolvedProvider,
-      log.resolvedModel,
-      log.credentialId,
-      log.attempts,
-      log.status,
-      log.errorCode,
-      log.inputTokens,
-      log.outputTokens,
-      log.cacheReadTokens,
-      log.cacheWriteTokens,
-      log.ttftMs,
-      log.durationMs,
-      log.costUsd,
-      JSON.stringify(log.degradations)
-    ]);
-    rollupLog(db, log);
+  const complete = db.transaction((log) => {
+    db.run(COMPLETE, values(log, "done"));
+    const stored = db.query("SELECT * FROM request_logs WHERE id = ?").get(log.id);
+    if (stored !== null)
+      rollupLog(db, toLog(stored));
   });
   return {
+    async begin(log) {
+      db.run(`INSERT INTO request_logs ${COLUMNS} VALUES ${PLACEHOLDERS}`, values(log, "pending"));
+    },
+    async route(id, target) {
+      db.run(`UPDATE request_logs
+            SET resolved_provider = ?, resolved_model = ?, credential_id = ?
+          WHERE id = ? AND state = 'pending'`, [target.provider, target.model, target.credentialId, id]);
+    },
     async append(log) {
-      insert(log);
+      complete(log);
+    },
+    async sweepPending() {
+      const stale = db.query("SELECT * FROM request_logs WHERE state = 'pending'").all().map(toLog);
+      for (const log of stale) {
+        complete({ ...log, status: 499, errorCode: "interrupted", durationMs: 0 });
+      }
+      return stale.length;
     },
     async recent(limit) {
       return db.query("SELECT * FROM request_logs ORDER BY at DESC LIMIT ?").all(limit).map(toLog);
@@ -22922,7 +22986,7 @@ function createUsageRepo(db) {
                   COALESCE(SUM(cache_write_tokens), 0) AS cache_write_tokens,
                   COALESCE(SUM(cost_usd), 0) AS cost_usd
              FROM ${daily ? "usage_daily" : "request_logs"}
-            WHERE ${timeColumn} >= ? AND ${timeColumn} <= ?
+            WHERE ${daily ? "" : "state = 'done' AND "}${timeColumn} >= ? AND ${timeColumn} <= ?
             GROUP BY key${split === null ? "" : ", split"}
             ORDER BY requests DESC`).all(since, until);
       return rows.map((r) => {
@@ -31467,17 +31531,17 @@ var ElysiaType = {
       elysiaMeta: "ArrayString"
     })).Decode((value) => {
       if (Array.isArray(value)) {
-        let values = [];
+        let values2 = [];
         for (let i = 0;i < value.length; i++) {
           const v = value[i];
           if (typeof v == "string") {
             const t2 = decode4(v, true);
-            Array.isArray(t2) ? values = values.concat(t2) : values.push(t2);
+            Array.isArray(t2) ? values2 = values2.concat(t2) : values2.push(t2);
             continue;
           }
-          values.push(v);
+          values2.push(v);
         }
-        return values;
+        return values2;
       }
       return typeof value == "string" ? decode4(value) : value;
     }).Encode((value) => {
@@ -31498,17 +31562,17 @@ var ElysiaType = {
       elysiaMeta: "ArrayQuery"
     })).Decode((value) => {
       if (Array.isArray(value)) {
-        let values = [];
+        let values2 = [];
         for (let i = 0;i < value.length; i++) {
           const v = value[i];
           if (typeof v == "string") {
             const t2 = decode4(v);
-            Array.isArray(t2) ? values = values.concat(t2) : values.push(t2);
+            Array.isArray(t2) ? values2 = values2.concat(t2) : values2.push(t2);
             continue;
           }
-          values.push(v);
+          values2.push(v);
         }
-        return values;
+        return values2;
       }
       return typeof value == "string" ? decode4(value) : value;
     }).Encode((value) => {
@@ -31552,16 +31616,16 @@ var ElysiaType = {
       sign
     }, v;
   },
-  UnionEnum: (values, options = {}) => {
-    const type = values.every((value) => typeof value == "string") ? { type: "string" } : values.every((value) => typeof value == "number") ? { type: "number" } : values.every((value) => value === null) ? { type: "null" } : {};
-    if (values.some((x) => typeof x == "object" && x !== null))
+  UnionEnum: (values2, options = {}) => {
+    const type = values2.every((value) => typeof value == "string") ? { type: "string" } : values2.every((value) => typeof value == "number") ? { type: "number" } : values2.every((value) => value === null) ? { type: "null" } : {};
+    if (values2.some((x) => typeof x == "object" && x !== null))
       throw new Error("This type does not support objects or arrays");
     return {
-      default: values[0],
+      default: values2[0],
       ...options,
       [Kind]: "UnionEnum",
       ...type,
-      enum: values
+      enum: values2
     };
   },
   NoValidate: (v, enabled = true) => (v.noValidate = enabled, v),
@@ -34007,33 +34071,33 @@ var getSchemaValidator = (s, {
         return v.then((v2) => Check22(v2, true));
       if (v.issues)
         return v;
-      const values = [];
-      return v && typeof v == "object" && values.push(v.value), runCheckers2(value, 0, values, v);
-    }, runCheckers2 = function(value, startIndex, values, lastV) {
+      const values2 = [];
+      return v && typeof v == "object" && values2.push(v.value), runCheckers2(value, 0, values2, v);
+    }, runCheckers2 = function(value, startIndex, values2, lastV) {
       for (let i = startIndex;i < checkers.length; i++) {
         let v = checkers[i].validate(value);
         if (v instanceof Promise)
           return v.then((resolved) => {
             if (resolved.issues)
               return resolved;
-            const nextValues = [...values];
+            const nextValues = [...values2];
             return resolved && typeof resolved == "object" && nextValues.push(resolved.value), runCheckers2(value, i + 1, nextValues, resolved);
           });
         if (v.issues)
           return v;
-        v && typeof v == "object" && values.push(v.value), lastV = v;
+        v && typeof v == "object" && values2.push(v.value), lastV = v;
       }
-      return mergeValues22(values, lastV);
-    }, mergeValues22 = function(values, lastV) {
-      if (!values.length)
+      return mergeValues22(values2, lastV);
+    }, mergeValues22 = function(values2, lastV) {
+      if (!values2.length)
         return { value: lastV };
-      if (values.length === 1)
-        return { value: values[0] };
-      if (values.length === 2)
-        return { value: mergeDeep(values[0], values[1]) };
-      let newValue = mergeDeep(values[0], values[1]);
-      for (let i = 2;i < values.length; i++)
-        newValue = mergeDeep(newValue, values[i]);
+      if (values2.length === 1)
+        return { value: values2[0] };
+      if (values2.length === 2)
+        return { value: mergeDeep(values2[0], values2[1]) };
+      let newValue = mergeDeep(values2[0], values2[1]);
+      for (let i = 2;i < values2.length; i++)
+        newValue = mergeDeep(newValue, values2[i]);
       return { value: newValue };
     };
     var Check2 = Check22, runCheckers = runCheckers2, mergeValues2 = mergeValues22;
@@ -38228,6 +38292,7 @@ async function dispatch(request2, deps, signal) {
   const deadlineAt = startedAt + snapshot.settings.requestDeadlineMs;
   const log = {
     id: crypto.randomUUID(),
+    state: "done",
     at: startedAt,
     apiKeyId: null,
     requestedModel: request2.model,
@@ -38319,6 +38384,11 @@ async function dispatch(request2, deps, signal) {
           log.credentialId = candidate.credential.id;
           log.resolvedProvider = candidate.target.provider;
           log.resolvedModel = candidate.target.model;
+          await deps.onRoute?.({
+            provider: candidate.target.provider,
+            model: candidate.target.model,
+            credentialId: candidate.credential.id
+          });
           log.inputTokens = 0;
           log.outputTokens = 0;
           log.cacheReadTokens = 0;
@@ -39116,18 +39186,44 @@ function parseOpenAIRequest(body2) {
 }
 
 // apps/gateway/src/logging.ts
+function report(what, requestId, error51) {
+  console.error(what, {
+    requestId,
+    reason: error51 instanceof Error ? error51.message : "unknown"
+  });
+}
+async function beginLog(store, log, keyId) {
+  try {
+    await store.usage.begin({ ...log, state: "pending", apiKeyId: keyId });
+  } catch (error51) {
+    report("failed to record request start", log.id, error51);
+  }
+}
+async function routeLog(store, requestId, target) {
+  try {
+    await store.usage.route(requestId, target);
+  } catch (error51) {
+    report("failed to record request route", requestId, error51);
+  }
+}
 async function finishLog(store, log, keyId) {
   try {
     await store.usage.append({ ...log, apiKeyId: keyId });
   } catch (error51) {
-    console.error("failed to persist request log", {
-      requestId: log.id,
-      reason: error51 instanceof Error ? error51.message : "unknown"
-    });
+    report("failed to persist request log", log.id, error51);
   }
 }
 
 // apps/gateway/src/routes/proxy.ts
+var KEEPALIVE_MS = 1e4;
+var KEEPALIVE = Symbol("keepalive");
+function withKeepalive(pending, ms) {
+  let timer;
+  const tick = new Promise((resolve) => {
+    timer = setTimeout(() => resolve(KEEPALIVE), ms);
+  });
+  return Promise.race([pending, tick]).finally(() => clearTimeout(timer));
+}
 var SSE_HEADERS = {
   "content-type": "text/event-stream; charset=utf-8",
   "cache-control": "no-cache, no-transform",
@@ -39146,7 +39242,7 @@ function asGatewayError2(error51) {
     return error51;
   return new GatewayError("INTERNAL", error51 instanceof Error ? error51.message : "internal error");
 }
-function sseResponse(frames, onDone) {
+function sseResponse(frames, onDone, keepaliveMs) {
   const encoder2 = new TextEncoder;
   let done = null;
   const runOnce = () => {
@@ -39154,10 +39250,19 @@ function sseResponse(frames, onDone) {
       done = onDone();
     return done;
   };
+  let inflight = null;
   const stream2 = new ReadableStream({
     async pull(controller) {
       try {
-        const next = await frames.next();
+        inflight ??= frames.next();
+        const next = await withKeepalive(inflight, keepaliveMs);
+        if (next === KEEPALIVE) {
+          controller.enqueue(encoder2.encode(`: keepalive
+
+`));
+          return;
+        }
+        inflight = null;
         if (next.done === true) {
           controller.close();
           await runOnce();
@@ -39182,7 +39287,9 @@ data: ${data}
 }
 async function handle(deps, rateLimiter, surface, request2) {
   const requestId = deps.requestId();
+  const startedAt = deps.now();
   let keyId = null;
+  let requestedModel = "";
   try {
     const key = await authenticateApiKey(deps.store, apiKeyHeader(request2.headers));
     keyId = key.id;
@@ -39192,11 +39299,42 @@ async function handle(deps, rateLimiter, surface, request2) {
     if (key.modelAllowlist !== null && !key.modelAllowlist.includes(chatRequest.model)) {
       throw new GatewayError("AUTH", `model "${chatRequest.model}" is not allowed for this API key`);
     }
-    const outcome = await dispatch(chatRequest, deps, request2.signal);
+    requestedModel = chatRequest.model;
+    let began = false;
+    const outcome = await dispatch(chatRequest, {
+      ...deps,
+      async onRoute(target) {
+        if (began) {
+          await routeLog(deps.store, requestId, target);
+          return;
+        }
+        began = true;
+        await beginLog(deps.store, {
+          id: requestId,
+          at: startedAt,
+          apiKeyId: keyId,
+          requestedModel,
+          resolvedProvider: target.provider,
+          resolvedModel: target.model,
+          credentialId: target.credentialId,
+          attempts: 0,
+          status: 0,
+          errorCode: null,
+          inputTokens: 0,
+          outputTokens: 0,
+          cacheReadTokens: 0,
+          cacheWriteTokens: 0,
+          ttftMs: null,
+          durationMs: 0,
+          costUsd: 0,
+          degradations: []
+        }, keyId);
+      }
+    }, request2.signal);
     const log = () => finishLog(deps.store, { ...outcome.log(), id: requestId }, keyId);
     if (chatRequest.stream) {
       const frames = surface === "anthropic" ? anthropicStream(outcome.events, requestId) : openaiStream(outcome.events, requestId, Math.floor(deps.now() / 1000));
-      return sseResponse(frames, log);
+      return sseResponse(frames, log, deps.keepaliveMs);
     }
     const events = [];
     for await (const event of outcome.events)
@@ -39215,9 +39353,10 @@ async function handle(deps, rateLimiter, surface, request2) {
     const gatewayError = asGatewayError2(error51);
     await finishLog(deps.store, {
       id: requestId,
-      at: deps.now(),
+      state: "done",
+      at: startedAt,
       apiKeyId: keyId,
-      requestedModel: "",
+      requestedModel,
       resolvedProvider: null,
       resolvedModel: null,
       credentialId: null,
@@ -39240,7 +39379,8 @@ function proxyRoutes(deps) {
   const rateLimiter = deps.rateLimiter ?? new ApiKeyRateLimiter(deps.now);
   const dispatchDeps = {
     ...deps,
-    snapshots: deps.snapshots ?? createRoutingSnapshotCache(deps.store)
+    snapshots: deps.snapshots ?? createRoutingSnapshotCache(deps.store),
+    keepaliveMs: deps.keepaliveMs ?? KEEPALIVE_MS
   };
   return new Elysia().post("/v1/messages", ({ request: request2 }) => handle(dispatchDeps, rateLimiter, "anthropic", request2)).post("/v1/chat/completions", ({ request: request2 }) => handle(dispatchDeps, rateLimiter, "openai", request2)).get("/v1/models", async ({ request: request2 }) => {
     try {
@@ -39469,6 +39609,9 @@ var store = await createStore({
   path: config2.databasePath,
   encryptionKey
 });
+var swept = await store.usage.sweepPending();
+if (swept > 0)
+  console.log(`retired ${swept} request(s) interrupted by the last shutdown`);
 var now = () => Date.now();
 var http = nodeHttpClient();
 var refresh = createRefresher({ store, providers: OAUTH_PROVIDERS, http, now });
@@ -39489,7 +39632,7 @@ var stopQuotaPoller = await startQuotaPoller({
   refresh,
   now
 });
-app.listen({ port: config2.port, hostname: config2.host });
+app.listen({ port: config2.port, hostname: config2.host, idleTimeout: 255 });
 console.log(`omnigateway listening on http://${config2.host}:${config2.port}`);
 var shuttingDown = false;
 function exitAfterClosingStore(code) {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "omnigateway",
-  "version": "0.1.3",
+  "version": "0.1.4",
   "description": "Self-hosted AI gateway with Anthropic- and OpenAI-compatible APIs, an admin console, and a CLI",
   "license": "MIT",
   "author": "Harismawan <mail@harismawan.com>",

package/public/assets/{Chip-D0YiqYkz.js → Chip-Ccqsb_Mq.js}

@@ -1,4 +1,4 @@
-import{B as e,Dt as t}from"./queries-vY5lJqBe.js";var n=t(),r={neutral:{fg:`var(--ink-dim)`,bg:`transparent`},ok:{fg:`var(--ok)`,bg:`var(--ok-wash)`},warn:{fg:`var(--warn)`,bg:`var(--warn-wash)`},down:{fg:`var(--down)`,bg:`var(--down-wash)`},accent:{fg:`var(--accent)`,bg:`var(--accent-wash)`}},i=e.span`
+import{Ot as e,V as t}from"./queries-BMwc6EvM.js";var n=e(),r={neutral:{fg:`var(--ink-dim)`,bg:`transparent`},ok:{fg:`var(--ok)`,bg:`var(--ok-wash)`},warn:{fg:`var(--warn)`,bg:`var(--warn-wash)`},down:{fg:`var(--down)`,bg:`var(--down-wash)`},accent:{fg:`var(--accent)`,bg:`var(--accent-wash)`}},i=t.span`
   display: inline-flex;
   align-items: center;
   gap: 4px;
@@ -12,7 +12,7 @@ import{B as e,Dt as t}from"./queries-vY5lJqBe.js";var n=t(),r={neutral:{fg:`var(
   font-size: 10.5px;
   letter-spacing: 0.02em;
   white-space: nowrap;
-`,a=e.span`
+`,a=t.span`
   display: inline-flex;
   align-items: center;
   gap: 5px;