Fix errors seen during longer runs: selenium exceptions, insecure certificates, recusrsion limit exceeded, BM25 document corpus too small

agent/tools/retreiveExamples.ts

@@ -15,6 +15,8 @@ const CACHE_PATH = "../data/csv.cache.json";
 
 const JSONL_PATH = "../data/input.jsonl"
 
+const BM25_MIN_DOCS = 3;
+
 type EmbeddingCache = {
   rawtexts: string[];
   cleantexts: string[];
@@ -287,8 +289,20 @@ async function embedText(text: string): Promise<number[]> {
 }
 
 function buildBM25(texts: string[]) {
-  logger.info("Building BM25 index (%s docs)...", texts.length);
+  let paddedTexts = texts;
 
+  if (texts.length < BM25_MIN_DOCS) {
+    const needed = BM25_MIN_DOCS - texts.length;
+    logger.error(
+      "Corpus too small for BM25 (%s docs, need %s+), padding with %s dummy doc(s)",
+      texts.length,
+      BM25_MIN_DOCS,
+      needed
+    );
+    paddedTexts = [...texts, ...Array(needed).fill("placeholder dummy document")];
+  }
+
+  logger.info("Building BM25 index (%s docs)...", paddedTexts.length);
   const bm25 = bm25Factory();
 
   bm25.defineConfig({
@@ -302,7 +316,7 @@ function buildBM25(texts: string[]) {
     nlp.tokens.removeWords,
   ]);
 
-  texts.forEach((text, i) => {
+  paddedTexts.forEach((text, i) => {
     bm25.addDoc({ text }, i);
   });
 

agent/tools/webpageFetch.ts

@@ -1,32 +1,92 @@
 import { Builder, Browser } from "selenium-webdriver";
 import firefox from "selenium-webdriver/firefox";
+import { backOff } from "exponential-backoff";
+import { logger } from "../utils/logger";
 
 export async function extractWebpageContent(url: string): Promise<string[]> {
+  try {
+    const response = await backOff(async () => {
+      return await extractWebpageContentWorker(url);
+    }, {
+      numOfAttempts: 10,
+      startingDelay: 500,
+      timeMultiple: 2,
+      jitter: "full",
+      maxDelay: 50000,
+    });
+    return response;
+  } catch (err: any) {
+    logger.error(`Failed out of retry loop for URL "${url}", returning placeholder to pipeline`);
+    return ["API EXCEPTION"];
+  }
+}
+
+async function extractWebpageContentWorker(url: string): Promise<string[]> {
+  let driver;
+  try {
     const options = new firefox.Options();
     options.addArguments("--headless");
+    driver = await new Builder()
+      .forBrowser(Browser.FIREFOX)
+      .setFirefoxOptions(options)
+      .build();
+  } catch (err: any) {
+    const desc = `Failed to launch Firefox driver: ${err.message}`;
+    logger.error(desc);
+    throw new Error(desc);
+  }
+
+  try {
+    try {
+      await driver.get(url);
+    } catch (err: any) {
+      const desc = `Failed to navigate to URL "${url}": ${err.message}`;
+      logger.error(desc);
+      throw new Error(desc);
+    }
 
-    let driver = await new Builder().forBrowser(Browser.FIREFOX).setFirefoxOptions(options).build()
     try {
-        await driver.get(url)
       await driver.wait(async () => {
         return await driver.executeScript(
           "return document.readyState === 'complete'"
         );
       }, 5000);
+    } catch (err: any) {
+      logger.error(`Page load timed out for "${url}", attempting to read partial content: ${err.message}`);
+      // do not throw, attempt to read
+    }
 
-        const readableText = await driver.executeScript(
+    let readableText: string;
+    try {
+      readableText = await driver.executeScript(
         "return document.body.innerText;"
       ) as string;
+    } catch (err: any) {
+      const desc = `Failed to extract page text from "${url}": ${err.message}`;
+      logger.error(desc);
+      throw new Error(desc);
+    }
 
     const filteredLines = readableText
       .split(/\r?\n/)
       .map(line => line.trim())
       .filter(line => line.split(/\s+/).length > 1);
 
+    if (filteredLines.length === 0) {
+      const desc = `No content extracted from "${url}"`;
+      logger.error(desc);
+      throw new Error(desc);
+    }
+
     return filteredLines;
   } finally {
-        await driver.quit()
+    try {
+      await driver.quit();
+    } catch (err: any) {
+      logger.error(`Failed to quit Firefox driver cleanly: ${err.message}`);
+    }
   }
 }
 
 // console.log(await extractWebpageContent("https://www.bbc.co.uk/news/live/c74wd01egvyt"))
+// console.log(await extractWebpageContent("https://badcertificate.int.jeynes.uk/"))

supporting/Wrapper/run.ts

@@ -118,7 +118,7 @@ async function processRecord(record: any): Promise<ResultRecord> {
       input: buildAgentInput(record),
       streamMode: "values",
       config: {
-        recursion_limit: 50
+        recursion_limit: 100
       }
     });