delivery optimization

updates

hel00

• directory-messages-ping [before: 1.79.27.0][after: 1.79.51.0] deployment
• manager-messages-ping [before: 1.110.3.0][after: 1.110.3.0] deployment
• go-ping [before: 1.18.11.0][after: 1.18.11.0] deployment
• directory-offline-package-sync [before: 1.79.27.0][after: 1.79.51.0] deployment
• subscribers-cache [before: 1.82.3.0][after: 1.79.51.0] deployment
• sender-events-proxy [before: 1.14.6.0][after: 1.79.51.0] deployment
• directory-messages [before: 1.82.3.0][after: 1.79.53.0] deployment

the rest of on-prems

• subscribers-cache [before: 1.82.3.0][after: 1.79.43.2] deployment

master

• directory-offline-package-uploader [before: 1.79.27.0][after: 1.79.51.0] deployment
• statistics-api [before: 1.23.1.0][after: 1.23.2.0] deployment
• callback-processor [before: 1.65.10.1][after: 1.79.51.0] deployment
• directory-events-listener [before: 1.52.2.1][after: 1.79.51.0] deployment
• directory-events-api [before: 1.81.9.0][after: 1.79.51.0] deployment
• directory-reach-detector [before: 1.52.2.1][after: 1.79.51.0] deployment
• directory-reach-extractor [before: 1.52.2.4][after: 1.79.51.0] deployment
• subscribers-tasks [before: 1.78.3.2][after: 1.79.43.1] deployment
• subscribers-errors [before: 1.82.3.0][after: 1.79.43.1] deployment
• di [before: 1.82.3.0][after: 1.79.52.0] deployment

deployment render all [[templates/items/deployment]] where page = @page.name

agenda

This deployment is meant to be performed on two clusters only:

• master (Amazon EKS)
• hel00 (Helsinki on-prem)

Deployment is only targeted on a single feature - delivery optimization. The feature is meant to improve the delivery-to-sent ratio while also decrease the amount of errors we see from Firebase (429). Feature consists of multiple parts:

• directory-reach-detector to detect reachability of subscribers
• directory-reach-extractor to extract reachability data from detector
• offline package manager (directory-offline-package-uploader) to look up for subscriber’s reachability (from directory-reach-extractor) and upload offline packages to S3
• directory-offline-package-sync to download ready offline packages from S3 and sync them to the directory on a slave cluster side, right next to the sender services
• directory-messages-ping and go-ping to ping subscribers marked within offline package
• callback-processor and subscribers-cache to trigger subscriber’s awake events for excluding them from the offline package
• statistics-api and clickhouse needs to be updated to support new data format for subscriber-sources-activities events

pre-deployment steps

Before doing anything foolish:

• make sure to make a backup of the mongo database on master cluster (subscribers_reach_data collection should be enough)