Page MenuHomePhabricator

Stress-test wdqs-categories
Open, Needs TriagePublic

Description

Currently, wdqs-categories is running on baremetal machines with 128 GB RAM, 32 or more cores, and RAID-0 backed SSDs. Before we move them to a separate environment (VMs or Kubernetes), we need to make sure the new environment has enough power to accommodate wdqs-categories.

Creating this ticket to:

  • Measure current usage in terms of HTTP interactions (hits per day, requests per second etc)
  • Devise a test that roughly approximates this level of usage (note that internal and external endpoints might not get the same level of traffic, so both should be tested)
  • Report the results

Event Timeline

bking updated the task description. (Show Details)

I haven't had much luck plumbing Turnilo or the SPARQL query rate Prometheus metric for Categories .

I did get a rough estimate via the nginx access logs on all wdqs hosts, using the command for n in $(ls /var/log/nginx/access.log.*.gz); do zgrep -v wmf-prometheus ${n}| grep -c "bigdata/namespace/categories" ; done' | tee categories-hits.out .

The internal hosts (which serve deepcat to the wikis) get far more traffic.

  • Internal: 18705 hits/day
  • External: 1477 hits/day
BTullis subscribed.

Removing this from the current milestone, as we are not currently working on the categories migration.