Page MenuHomePhabricator

Near zero downtime Data reload for WCQS
Closed, ResolvedPublic5 Estimated Story Points

Description

As a user I want a shortest possible downtime for WCQS in case of data reload, so that I can continue using that.

Proposed solutions:

  • Feed the second blazegraph instance, switch journals afterwards
  • The same but with two namespaces (single instance of Blazegraph)

Category reload happens this way already.

Switch can also happen via nginx.

AC:

  • data reload doesn't trigger maintenance page
  • data is available after reload.

Event Timeline

Restricted Application added a subscriber: Aklapper. · View Herald Transcript

In WDQS, there are several instances and they are not reloaded at the same time.

Using a single instance with 2 namespaces seems to be the faster way to implement this.

Gehel set the point value for this task to 5.Sep 14 2020, 5:31 PM
Zbyszko triaged this task as High priority.

Change 627485 had a related patch set uploaded (by ZPapierski; owner: ZPapierski):
[wikidata/query/rdf@master] Provide active/standby indexes for WCQS

https://gerrit.wikimedia.org/r/627485

Change 627495 had a related patch set uploaded (by ZPapierski; owner: ZPapierski):
[operations/puppet@production] Switch between active W[D|C]QS indexes

https://gerrit.wikimedia.org/r/627495

Change 627495 abandoned by ZPapierski:
[operations/puppet@production] Switch between active W[D|C]QS indexes

Reason:
Made obsolete.

https://gerrit.wikimedia.org/r/627495

Change 627485 merged by jenkins-bot:
[wikidata/query/rdf@master] Provide active/standby namespaces for WCQS

https://gerrit.wikimedia.org/r/627485