You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
In the new govuk index, an unpublishing (non-withdawn) message causes rummager to delete the document from the index. This can lead to a race condition if a publishing and unpublishing message arrive out of order.
If a publisher publishes and then immediately unpublishes a document, but the unpublishing message arrives first, the document will be removed from the index. Then the stale publishing message arrives, so rummager will incorrectly re-add the document to the index.
We could avoid this by adding temporary "tombstone" records when a document is deleted, i.e. keep the document in the index with a flag marking it as unpublished. The document will be removed when the index is next recreated, but while it exists it would prevent the document from being republished because Elasticsearch's versioning would cause the stale write to fail.
The text was updated successfully, but these errors were encountered:
From https://trello.com/c/ap5pQF5R/20-add-tombstone-records-for-deleted-documents
In the new govuk index, an unpublishing (non-withdawn) message causes rummager to delete the document from the index. This can lead to a race condition if a publishing and unpublishing message arrive out of order.
If a publisher publishes and then immediately unpublishes a document, but the unpublishing message arrives first, the document will be removed from the index. Then the stale publishing message arrives, so rummager will incorrectly re-add the document to the index.
We could avoid this by adding temporary "tombstone" records when a document is deleted, i.e. keep the document in the index with a flag marking it as unpublished. The document will be removed when the index is next recreated, but while it exists it would prevent the document from being republished because Elasticsearch's versioning would cause the stale write to fail.
The text was updated successfully, but these errors were encountered: