Record feedback target author and allow filtering by feedback authors

[MrOrz]

Fixes #280

• Update schema and expose replyUserId, articleReplyUserId for ArticleReplyFeedback
• CreateOrUpdateArticleReplyFeedback writes replyUserId and articleReplyUserId on creation
• ListArticleReplyFeedback can filter by replyUserId, articleReplyUserId or authorId
  • authorId: lists feedbacks whose replyUserId is the specified ID, or whose articleReplyUserId is the specified ID
• Migration script filling up replyUserId and articleReplyUserId for existing ArticleReplyFeedback (not checked in, just for copy & paste use)

[coveralls]

Coverage increased (+0.07%) to 87.422% when pulling 8b5981141f81fae7ddace91c6efed553efe3abc5 on record-feedback-target-author into d806fc6f179aa68caf13f0b9cc54ebfc5f52cb37 on master.

[MrOrz]

Migration script (not checked in repo)

// src/scripts/migrations/fillFeedbackAuthors.js
/**
 * How to use this script to migrate elasticsearch schema
 *  1. First of all, ensure latest API has been deployed; new feedbacks have `replyUserId` and `articleReplyUserId` filled.
 *  2. On target machine, update schema and reindex the following indexes using `npm run reload -- <index name>`:
 *     analytics (optional), articlereplyfeedbacks
 *  1. On local machine, put script to under rumors-api project directory's src/scripts
 *  2. Start ssh port forwarding `ssh -L 62222:staging.server.path.to:62222 root@staging.server.path.to`
 *  3. Add `ELASTICSEARCH_URL=http://localhost:62222` to .env
 *  4. Run `npx babel-node src/scripts/migrations/fillFeedbackAuthors.js`
 */
import 'dotenv/config';
import { SingleBar } from 'cli-progress';

import getAllDocs from 'util/getAllDocs';
import client from 'util/client';
import DataLoaders from 'graphql/dataLoaders';

// query to list out article reply feedbacks without replyUserId
const QUERY_TO_PROCESS = {
  bool: {
    must_not: {
      exists: { field: 'replyUserId' },
    },
  },
};

async function processBatch(feedbackMap) {
  let operations = [];
  for (const [_id, doc] of feedbackMap) {
    operations.push(
      { update: { _index: 'articlereplyfeedbacks', _type: 'doc', _id } },
      { doc }
    );
  }
  await client.bulk({ body: operations });
}

async function main() {
  const loader = new DataLoaders();
  const bar = new SingleBar({ stopOnComplete: true });

  const {
    body: { count },
  } = await client.count({
    index: 'articlereplyfeedbacks',
    type: 'doc',
    body: {
      query: QUERY_TO_PROCESS,
    },
  });

  bar.start(count, 0);

  // Feedback ID to doc
  let feedbackMap = new Map();
  let promises = [];

  for await (const { _id: id, _source: feedback } of getAllDocs(
    'articlereplyfeedbacks',
    QUERY_TO_PROCESS,
    { size: 100 }
  )) {
    bar.increment();
    promises.push(
      loader.docLoader
        .loadMany([
          {
            index: 'replies',
            id: feedback.replyId,
          },
          {
            index: 'articles',
            id: feedback.articleId,
          },
        ])
        .then(([{ userId: replyUserId }, article]) => {
          const articleReply = article.articleReplies.find(
            ar => ar.replyId === feedback.replyId
          );
          if (!articleReply) return;

          const { userId: articleReplyUserId } = articleReply;
          feedbackMap.set(id, { replyUserId, articleReplyUserId });
        })
    );

    if (promises.length === 100) {
      await Promise.all(promises);
      promises = [];

      await processBatch(feedbackMap);
      feedbackMap = new Map();
    }
  }

  await Promise.all(promises);
  await processBatch(feedbackMap);

  bar.stop();
}

if (require.main === module) {
  main();
}

Test target:

• 2022-04-XX production snapshot on local machine
• 304,176 article-reply-feedbacks to update

Run result:

• Takes 1~2 minutes to reindex
• Takes about an hour to process
• Encounter 134 feedbacks to non-exist article-reply, probably caused by:
  0. the user initiates a search session by sending a message that causes chatbot to return multiple matching articles
  1. the user clicks one article, its reply and feedback buttons shows
  2. the user scrolls back and clicks another article, its reply and feedback button shows -- meanwhile the context's article ID has changed
  3. the user scrolls back again and react on the old feedback button within the same search session -- the reaction is not blocked because it's within the same search session, but the article ID in the context (the new one) does not match the reply ID for the button (the old one)

---

Staging run result

• 2022-06-19 production snapshot; 317,615 feedbacks to process
• Reindexed from articlereplyfeedbacks_v1_1_1 to articlereplyfeedbacks_v1_2_0 in 36 seconds.
• Reindexed from analytics_v1_0_2 to analytics_v1_2_0 in 576 seconds.
• Takes 16min to migrate (v2 script)
• 2844 feedbacks left, who cannot be migrated (probably due to incorrect feedback articleId / replyId)

[MrOrz]

Migration report on production

• Migration: 2022/7/1 1:11:53 ~ 1:26:13
• Feedbacks matching criteria: 320,260
• Feedbacks cannot be migrated: 2,848