25-Aug-2026 — keeping the right version live, and a few quieter fixes
None of this changes how you run a walk or send an email. It is the late-August work that sits underneath the sites.
On 23 August a six-month-old version of the site went out to every group. These notes cover the follow-up that stops that happening the same way again, plus a few quieter fixes: false "the site is down" alarms while a new version was going out, secret keys that could appear in error logs, a fiddly admin screen for shared-service logins, and already-read inbox conversations bouncing back to unread after a restart.
At a glance
- We can no longer accidentally put an old version of the site on every group, which is what happened on 23 August
- After an update, each site has to prove it is running the version we asked for
- A check runs around the clock so we notice if a site is down, and it stays quiet while a new version is going out so restarting machines are not reported as down
- Error logs from talking to Head office about supporter records can no longer include the secret key used for that call
- Restarting the site no longer marks inbox conversations you have already read as unread
- The admin screen that holds logins for shared services (mail, file storage and the like) is easier to read
Don't raise the alarm while a new version is going out
When we put a new version of the site on the groups, the machines briefly restart. A scheduled "are the sites up?" check was firing during that restart and sending false alarms. It now waits until the update has finished.
Stop putting the wrong version of the site on every group #362
On 23 August a six-month-old version of the site went out to every group. The number we used to pick "which version" had been read from the wrong list of jobs, so we thought we were sending the new one and sent February's instead.
That cannot happen the same way again. We now check the version number is a real, recent build before anything goes out. After each site updates, it has to prove it is serving that version. And a check runs around the clock so we notice if a site is down.
You can still deliberately put an older version back if you mean to.
Keep secret keys out of the logs #328
When NGX talks to Head office about supporter records, the call uses a secret key. If that call failed, the error written to the logs could include the key. The logs now strip it out every time, and tests prove they do.
There was also leftover data from an older way of downloading those supporter records, which remembered how far the last download had got. Nothing used it any more, so it has been removed.
A clearer screen for shared-service logins #287
This is the admin screen that holds the logins for services used across the sites (mail, file storage and the like). Each service now sits on its own card, instead of being packed into a long list. You can show only the settings that apply everywhere, only the ones that differ per site, or both. If you switch which site you are looking at, the service you had selected stays selected. When several sites share the same value, that is marked on the card.
Keep inbox conversations marked as read after a restart #278
When the site restarts, it tidies inbox conversations so that messages you sent from the group address sit in the right place. That tidy-up used to look only at whether the last message was incoming or outgoing, and could mark a conversation you had already read as unread again. It now respects that you have already read it.