Page MenuHomePhabricator

Tune WDQS backend storage params, and document the decision
Closed, ResolvedPublic1 Estimated Story Points

Description

Now that we can updating the index more regularly and that the qlever index rebuilds are released, we have a better idea of what the storage needs are for the system.

We ran into an issue where there wasn't room to download a third index given the delta-triples plus two existing indices in a 600GiB volume. There might be additional issues like this, but nothing obvious comes to mind now so we will file follow up tickets if they arise.

Acceptance criterion:

  • document why 900 GiB in the doc mentioned here: T438165

Event Timeline

Change #1329388 had a related patch set uploaded (by Lerickson; author: Lerickson):

[operations/deployment-charts@master] Increase WDQS backend pod storage to 900GiB.

https://gerrit.wikimedia.org/r/1329388

Change #1329388 merged by jenkins-bot:

[operations/deployment-charts@master] Increase WDQS backend pod storage to 900GiB.

https://gerrit.wikimedia.org/r/1329388

trueg renamed this task from [NEEDS GROOMING] Tune WDQS backend storage and memory params to Tune WDQS backend storage and memory params.Sep 2 2026, 2:04 PM
trueg renamed this task from Tune WDQS backend storage and memory params to Tune WDQS backend storage params.Sep 2 2026, 2:10 PM
trueg updated the task description. (Show Details)
trueg changed the status of subtask T436301: wdqs backend pods: upsize topoLVM from 600GiB->900GiB from Open to In Progress.
trueg renamed this task from Tune WDQS backend storage params to [NEEDS GROOMING] Tune WDQS backend storage params.Sep 11 2026, 11:04 AM
trueg removed trueg as the assignee of this task.
trueg subscribed.
lerickson renamed this task from [NEEDS GROOMING] Tune WDQS backend storage params to Tune WDQS backend storage params, and document the decision.Wed, Sep 16, 2:00 PM
lerickson updated the task description. (Show Details)
lerickson set the point value for this task to 1.

document why 900 GiB in the doc mentioned here: T438165

@lerickson IIRC the AC was to document the adoption of TopoLVM in a decision log (done here ). For the sizing itself, I think we properly documented in the right place (where the change is applied).

Currently we have 2TBs per k8s node. Each node has two PODs deployed (main and scholarly), so 900GB gives enough headroom for updating the indexes and persisting delta triples) of both PODs.

LGTM @gmodena , thanks. Agreed that 900 GiB can be documented at the source. However, I think it still says this is based on 3 indices (1 live, 1 backup, 1 being downloaded), whereas we'd decided to have no backup. Am I misremembering?
https://gerrit.wikimedia.org/r/plugins/gitiles/operations/deployment-charts/+/refs/heads/master/helmfile.d/dse-k8s-services/_wdqs_common_/values.yaml#7