Keyboard shortcuts

Press ← or → to navigate between chapters

Press S or / to search in the book

Press ? to show this help

Press Esc to hide this help

Search Personalization/Metarank Deployment Plan

QA Test Staging

https://hungryhubgroup.slack.com/archives/C01C1U5LBFT/p1746504212204959?thread_ts=1746412210.398219&cid=C01C1U5LBFT

Terraform

ISSUE: It works fine when accessing it from a pod/service using port forwarding, but it does not work when accessing it using the domain https://metarank.hhstaging.dev . Maybe the issue is with the load balancer. Please help resolve it, Mas @Irfan Wicaksana.

Setup Kafka @Luthfi

  • Create topics
    • Engineering
    • Prod Support
    • Production
  • Setup S3 Sink Connector
    • Engineering
    • Staging (no need)
    • Production

Jupyter Notebook @Luthfi

  • Exec create s3 shell (cancelled, use terraform instead)

HH-Personalize Server @Luthfi

  • Review PR
  • Release the PR to main
  • Monitor & Check
    • Server
    • BullMQ
    • Web
    • IOS
    • Android

HH-Search Server @Luthfi

  • Add METARANK_URL env
  • Review PR
  • Release the PR to main
  • Monitor
    • Server
    • BullMQ
    • Web
    • IOS
    • Android

Check & Re-check @Luthfi

Note: waiting for the metarank release to prod

  • Health check metarank
    • https://metarank.hungryhub.com/health
    • https://metarank.hungryhub.com/metrics

HH-Pegasus @afif

Don’t release until all the above tasks are completed, but you can review it first.

Note: waiting for HH-Pegasus release to prod

  • BullMQ check events https://personalize.hungryhub.com/bullmq
    • Impression Event
    • Interaction Event

Post Release @Luthfi

  • Disable Impression & Interaction event worker
  • Load restaurants data
  • Load users data
  • Enable Impression & Interaction event worker
  • Wait the data collection for 1 or 1/2 day for initial training

NOTE: To train Metarank, impression events are necessary. Currently, we do not have these events, so we need to collect them before train the model.

  • Validate the training data
    • Pull the data from s3 using jupyter notebook
    • Validate the data
  • Train the model (try in local standalone cmd first)
  • Train the model (production)
  • Test the API using production build preview
  • Enable the Growthbook Feature Flag

https://growthbook.hungryhub.com/features/search-ranking

  • All users with “@hungryhub.com” email addresses will always get personalized search results.

  • Then, we’ll gradually roll out to other users, starting with 5% and gradually increasing while monitoring the service.

Live Review QA

Attachments