(core) External requests

Summary:
Adds a Python function `REQUEST` which makes an HTTP GET request. Behind the scenes it:

- Raises a special exception to stop trying to evaluate the current cell and just keep the existing value.
- Notes the request arguments which will be returned by `apply_user_actions`.
- Makes the actual request in NodeJS, which sends back the raw response data in a new action `RespondToRequests` which reevaluates the cell(s) that made the request.
- Wraps the response data in a class which mimics the `Response` class of the `requests` library.

In certain cases, this asynchronous flow doesn't work and the sandbox will instead synchronously call an exported JS method:

- When reevaluating a single cell to get a formula error, the request is made synchronously.
- When a formula makes multiple requests, the earlier responses are retrieved synchronously from files which store responses as long as needed to complete evaluating formulas. See https://grist.slack.com/archives/CL1LQ8AT0/p1653399747810139

Test Plan: Added Python and nbrowser tests.

Reviewers: georgegevoian

Reviewed By: georgegevoian

Subscribers: paulfitz, dsagal

Differential Revision: https://phab.getgrist.com/D3429

app/server/lib/ActiveDoc.ts

@@ -4,7 +4,13 @@
  * change events.
  */
 
-import {getEnvContent, LocalActionBundle, SandboxActionBundle, UserActionBundle} from 'app/common/ActionBundle';
+import {
+  getEnvContent,
+  LocalActionBundle,
+  SandboxActionBundle,
+  SandboxRequest,
+  UserActionBundle
+} from 'app/common/ActionBundle';
 import {ActionGroup, MinimalActionGroup} from 'app/common/ActionGroup';
 import {ActionSummary} from "app/common/ActionSummary";
 import {
@@ -33,12 +39,7 @@ import {
   UserAction
 } from 'app/common/DocActions';
 import {DocData} from 'app/common/DocData';
-import {
-  getDataLimitRatio,
-  getDataLimitStatus,
-  getSeverity,
-  LimitExceededError,
-} from 'app/common/DocLimits';
+import {getDataLimitRatio, getDataLimitStatus, getSeverity, LimitExceededError} from 'app/common/DocLimits';
 import {DocSnapshots} from 'app/common/DocSnapshot';
 import {DocumentSettings} from 'app/common/DocumentSettings';
 import {
@@ -75,6 +76,7 @@ import {GRIST_DOC_SQL, GRIST_DOC_WITH_TABLE1_SQL} from 'app/server/lib/initialDo
 import {ISandbox} from 'app/server/lib/ISandbox';
 import * as log from 'app/server/lib/log';
 import {LogMethods} from "app/server/lib/LogMethods";
+import {DocRequests} from 'app/server/lib/Requests';
 import {shortDesc} from 'app/server/lib/shortDesc';
 import {TableMetadataLoader} from 'app/server/lib/TableMetadataLoader';
 import {DocTriggers} from "app/server/lib/Triggers";
@@ -182,6 +184,7 @@ export class ActiveDoc extends EventEmitter {
 
   private _log = new LogMethods('ActiveDoc ', (s: OptDocSession) => this.getLogMeta(s));
   private _triggers: DocTriggers;
+  private _requests: DocRequests;
   private _dataEngine: Promise<ISandbox>|undefined;
   private _activeDocImport: ActiveDocImport;
   private _onDemandActions: OnDemandActions;
@@ -270,6 +273,7 @@ export class ActiveDoc extends EventEmitter {
     this.docStorage = new DocStorage(docManager.storageManager, docName);
     this.docClients = new DocClients(this);
     this._triggers = new DocTriggers(this);
+    this._requests = new DocRequests(this);
     this._actionHistory = new ActionHistoryImpl(this.docStorage);
     this.docPluginManager = new DocPluginManager(docManager.pluginManager.getPlugins(),
       docManager.pluginManager.appRoot!, this, this._docManager.gristServer);
@@ -1095,7 +1099,7 @@ export class ActiveDoc extends EventEmitter {
       this.dataLimitStatus === "deleteOnly" &&
       !actions.every(action => [
           'RemoveTable', 'RemoveColumn', 'RemoveRecord', 'BulkRemoveRecord',
-          'RemoveViewSection', 'RemoveView', 'ApplyUndoActions',
+          'RemoveViewSection', 'RemoveView', 'ApplyUndoActions', 'RespondToRequests',
         ].includes(action[0] as string))
     ) {
       throw new Error("Document is in delete-only mode");
@@ -1420,6 +1424,10 @@ export class ActiveDoc extends EventEmitter {
       }
       const user = docSession ? await this._granularAccess.getCachedUser(docSession) : undefined;
       sandboxActionBundle = await this._rawPyCall('apply_user_actions', normalActions, user?.toJSON());
+      const {requests} = sandboxActionBundle;
+      if (requests) {
+        this._requests.handleRequestsBatchFromUserActions(requests).catch(e => console.error(e));
+      }
       await this._reportDataEngineMemory();
     } else {
       // Create default SandboxActionBundle to use if the data engine is not called.
@@ -2087,6 +2095,7 @@ export class ActiveDoc extends EventEmitter {
       preferredPythonVersion,
       sandboxOptions: {
         exports: {
+          request: (key: string, args: SandboxRequest) => this._requests.handleSingleRequestWithCache(key, args),
           guessColInfo: (values: Array<string | null>) =>
             guessColInfoWithDocData(values, this.docData!),
           convertFromColumn: (...args: Parameters<ReturnType<typeof convertFromColumn>>) =>

app/server/lib/GranularAccess.ts

@@ -4,7 +4,13 @@ import { ActionGroup } from 'app/common/ActionGroup';
 import { createEmptyActionSummary } from 'app/common/ActionSummary';
 import { ServerQuery } from 'app/common/ActiveDocAPI';
 import { ApiError } from 'app/common/ApiError';
-import { AddRecord, BulkAddRecord, BulkColValues, BulkRemoveRecord, BulkUpdateRecord } from 'app/common/DocActions';
+import {
+  AddRecord,
+  BulkAddRecord,
+  BulkColValues,
+  BulkRemoveRecord,
+  BulkUpdateRecord,
+} from 'app/common/DocActions';
 import { RemoveRecord, ReplaceTableData, UpdateRecord } from 'app/common/DocActions';
 import { CellValue, ColValues, DocAction, getTableId, isSchemaAction } from 'app/common/DocActions';
 import { TableDataAction, UserAction } from 'app/common/DocActions';

app/server/lib/Requests.ts

@@ -0,0 +1,128 @@
+import {SandboxRequest} from 'app/common/ActionBundle';
+import {ActiveDoc} from 'app/server/lib/ActiveDoc';
+import {makeExceptionalDocSession} from 'app/server/lib/DocSession';
+import {httpEncoding} from 'app/server/lib/httpEncoding';
+import {HttpsProxyAgent} from 'https-proxy-agent';
+import {HttpProxyAgent} from 'http-proxy-agent';
+import fetch from 'node-fetch';
+import * as path from 'path';
+import * as tmp from 'tmp';
+import chunk = require('lodash/chunk');
+import fromPairs = require('lodash/fromPairs');
+import zipObject = require('lodash/zipObject');
+import * as fse from 'fs-extra';
+import * as log from 'app/server/lib/log';
+
+export class DocRequests {
+  // Request responses are briefly cached in files only to handle multiple requests in a formula
+  // and only as long as needed to finish calculating all formulas.
+  // When _numPending reaches 0 again, _cacheDir is deleted.
+  private _numPending: number = 0;
+  private _cacheDir: tmp.SynchrounousResult | null = null;
+
+  constructor(private readonly _activeDoc: ActiveDoc) {}
+
+  public async handleRequestsBatchFromUserActions(requests: Record<string, SandboxRequest>) {
+    const numRequests = Object.keys(requests).length;
+    this._numPending += numRequests;
+    try {
+      // Perform batches of requests in parallel for speed, and hope it doesn't cause rate limiting...
+      for (const keys of chunk(Object.keys(requests), 10)) {
+        const responses: Response[] = await Promise.all(keys.map(async key => {
+          const request = requests[key];
+          const response = await this.handleSingleRequestWithCache(key, request);
+          return {
+            ...response,
+            // Tells the engine which cell(s) made the request and should be recalculated to use the response
+            deps: request.deps,
+          };
+        }));
+        // Tell the sandbox which previous responses we have cached in files.
+        // This lets it know it can immediately and synchronously get those responses again.
+        const cachedRequestKeys = await fse.readdir(this._cacheDir!.name);
+        // Recalculate formulas using this batch of responses.
+        const action = ["RespondToRequests", zipObject(keys, responses), cachedRequestKeys];
+        await this._activeDoc.applyUserActions(makeExceptionalDocSession("system"), [action]);
+      }
+    } finally {
+      this._numPending -= numRequests;
+      if (this._numPending === 0) {
+        log.debug(`Removing DocRequests._cacheDir: ${this._cacheDir!.name}`);
+        this._cacheDir!.removeCallback();
+        this._cacheDir = null;
+      }
+    }
+  }
+
+  public async handleSingleRequestWithCache(key: string, request: SandboxRequest): Promise<Response> {
+    if (!this._cacheDir) {
+      // Use the sync API because otherwise multiple requests being handled at the same time
+      // all reach this point, `await`, and create different dirs.
+      // `unsafeCleanup: true` means the directory can be deleted even if it's not empty, which is what we expect.
+      this._cacheDir = tmp.dirSync({unsafeCleanup: true});
+      log.debug(`Created DocRequests._cacheDir: ${this._cacheDir.name}`);
+    }
+
+    const cachePath = path.resolve(this._cacheDir.name, key);
+    try {
+      const result = await fse.readJSON(cachePath);
+      result.content = Buffer.from(result.content, "base64");
+      return result;
+    } catch {
+      const result = await this._handleSingleRequestRaw(request);
+      const resultForJson = {...result} as any;
+      if ('content' in result) {
+        resultForJson.content = result.content.toString("base64");
+      }
+      fse.writeJSON(cachePath, resultForJson).catch(e => log.warn(`Failed to save response to cache file: ${e}`));
+      return result;
+    }
+  }
+
+  private async _handleSingleRequestRaw(request: SandboxRequest): Promise<Response> {
+    try {
+      if (process.env.GRIST_EXPERIMENTAL_PLUGINS != '1') {
+        throw new Error("REQUEST is not enabled");
+      }
+      const {url, params, headers} = request;
+      const urlObj = new URL(url);
+      log.rawInfo("Handling sandbox request", {host: urlObj.host, docId: this._activeDoc.docName});
+      for (const [param, value] of Object.entries(params || {})) {
+        urlObj.searchParams.append(param, value);
+      }
+      const response = await fetch(urlObj.toString(), {headers: headers || {}, agent: proxyAgent(urlObj)});
+      const content = await response.buffer();
+      const {status, statusText} = response;
+      const encoding = httpEncoding(response.headers.get('content-type'), content);
+      return {
+        content, status, statusText, encoding,
+        headers: fromPairs([...response.headers]),
+      };
+    } catch (e) {
+      return {error: String(e)};
+    }
+  }
+}
+
+interface SuccessfulResponse {
+  content: Buffer;
+  status: number;
+  statusText: string;
+  encoding?: string;
+  headers: Record<string, string>;
+}
+
+interface RequestError {
+  error: string;
+}
+
+type Response = RequestError | SuccessfulResponse;
+
+function proxyAgent(requestUrl: URL) {
+  const proxy = process.env.GRIST_HTTPS_PROXY;
+  if (!proxy) {
+    return undefined;
+  }
+  const ProxyAgent = requestUrl.protocol === "https:" ? HttpsProxyAgent : HttpProxyAgent;
+  return new ProxyAgent(proxy);
+}

app/server/lib/Sharing.ts

@@ -6,7 +6,7 @@ import {
   LocalActionBundle,
   UserActionBundle
 } from 'app/common/ActionBundle';
-import {DocAction, getNumRows, UserAction} from 'app/common/DocActions';
+import {CALCULATING_USER_ACTIONS, DocAction, getNumRows, UserAction} from 'app/common/DocActions';
 import {allToken} from 'app/common/sharing';
 import * as log from 'app/server/lib/log';
 import {LogMethods} from "app/server/lib/LogMethods";
@@ -215,8 +215,7 @@ export class Sharing {
 
     try {
 
-      const isCalculate = (userActions.length === 1 &&
-                           (userActions[0][0] === 'Calculate' || userActions[0][0] === 'UpdateCurrentTime'));
+      const isCalculate = (userActions.length === 1 && CALCULATING_USER_ACTIONS.has(userActions[0][0] as string));
       // `internal` is true if users shouldn't be able to undo the actions. Applies to:
       // - Calculate/UpdateCurrentTime because it's not considered as performed by a particular client.
       // - Adding attachment metadata when uploading attachments,

app/server/lib/httpEncoding.ts

@@ -0,0 +1,43 @@
+// Based on the source code of the Body.textConverted method in node-fetch
+export function httpEncoding(header: string | null, content: Buffer): string | undefined {
+  let res: RegExpExecArray | null = null;
+
+  // header
+  if (header) {
+    res = /charset=([^;]*)/i.exec(header);
+  }
+
+  // no charset in content type, peek at response body for at most 1024 bytes
+  const str = content.slice(0, 1024).toString();
+
+  // html5
+  if (!res && str) {
+    res = /<meta.+?charset=(['"])(.+?)\1/i.exec(str);
+  }
+
+  // html4
+  if (!res && str) {
+    res = /<meta\s+?http-equiv=(['"])content-type\1\s+?content=(['"])(.+?)\2/i.exec(str);
+
+    if (res) {
+      res = /charset=(.*)/i.exec(res.pop()!);
+    }
+  }
+
+  // xml
+  if (!res && str) {
+    res = /<\?xml.+?encoding=(['"])(.+?)\1/i.exec(str);
+  }
+
+  // found charset
+  if (res) {
+    let charset = res.pop();
+
+    // prevent decode issues when sites use incorrect encoding
+    // ref: https://hsivonen.fi/encoding-menu/
+    if (charset === 'gb2312' || charset === 'gbk') {
+      charset = 'gb18030';
+    }
+    return charset;
+  }
+}