Page MenuHomePhabricator

Re-structure WME S3 key space design to avoid hot partitions issue
Closed, ResolvedPublic3 Estimated Story Points

Description

Avoid s3 hot partition issues:
Existing prefix: articles_v2/enwiki/The_Answer_to_Both_Your_Questions.json
Extended load on s3 causes partitioning, thus increases throughput but having the articles directly under enwiki leads to excessive throttling because s3 does not partition them well compared to articles_v2/enwiki/b/b1/The_Answer_to_Both_Your_Questions.json
This follows the WMF’s nomenclature to avoid the same issue.

  • Design doc for this ticket
  • Ensure the files are stored in s3 following the agreed structure
  • Adapt APIs to fetch files using this structure

Event Timeline

We need input from Renil, the agreed to break the ticket into 2: scoping and implementation

@KMontalva-WMF can you please add the work you are doing on design work to this ticket?

I'm not doing documentation, I wrote a design doc for this ticket.

KMontalva-WMF set the point value for this task to 3.

I'm opening this one just for tracking purposes, I'll close it tomorrow again