tl;dr: scratch volumes were broken for about 45 minutes. Things are now resolved or resolving, and no action is needed on your part.

--

The transfer to the new scratch server did not go as smoothly as I hoped. Any toolforge jobs or VM processes that accessed the scratch volume were probably hanging or crashing for about 45 minutes, from around 15:15 UTC to 16:00 UTC.

This has now been resolved on all cloud-vps hosts, but some toolforge jobs are probably still stuck. We are gradually rescheduling all pods, so no action is needed on your part other than a bit more waiting.

More details can be found on https://phabricator.wikimedia.org/T433426 ; eventually a full incident report will be published and linked to from that task.

-Andrew



On 7/28/26 3:17 PM, Andrew Bogott wrote:
This email relates to the shared /data/scratch mount used by toolforge and some cloud-vps projects. It is provided to the following projects:

- dumps
- language
- maps
- math
- mwoffliner
- paws
- statanalyser
- testlabs
- tools
- toolsbeta
- video
- wikidata-query
- wikidumpparse
- wikiqlever

As part of our push to eliminate the use of Debian Bullseye, we will be replacing the server that hosts the nfs scratch volume on Monday, sometime around 15:00 UTC. Testing suggests that this will only produce a few minutes of downtime for the volume (likely reported as 'Permission Denied'); if you find your system frozen please give it a few minutes and then reboot if waiting doesn't help.

The data on the old server will be transferred to the new server, so no data will be lost.[0]

Bonus points: If you are a cloud-vps admin and have read this far, please consider whether you still need the scratch volume. This service was initially provided before we had attachable block storage available[1], so many users rely on scratch not for file transfer between hosts but because they needed extra storage space. If you are one of those cases, please open a quota ticket[2] and mention that you are moving off of scratch. Life without NFS is better for everyone!

This work is tracked on phabricator here: https://phabricator.wikimedia.org/T433426

- Andrew + the WMCS team

[0] ...but if you have data on scratch that you care about, it should be somewhere else! We make no reliability guarantees for this storage. [1] https://wikitech.wikimedia.org/wiki/Help:Adding_disk_space_to_Cloud_VPS_instances
[2] https://phabricator.wikimedia.org/project/view/2880/


_______________________________________________
Cloud-announce mailing list -- [email protected]
List information: 
https://lists.wikimedia.org/postorius/lists/cloud-announce.lists.wikimedia.org/

Reply via email to