Imports and copies no longer run on the part of the portal that routes traffic

Inside a portal, one component acts as the switchboard: every message anyone sends passes through it on the way to whoever should handle it. It is deliberately kept free — a switchboard that stops to do a job of its own is a switchboard nobody's call gets through.

Some of the portal's own background work was, in effect, calling from the switchboard. A bulk import, a node copy, a move — each was written to run wherever its caller happened to be standing, and several of the callers stand on the switchboard: the content importers, the parts that run before anyone has logged in, the plumbing that sets up a guest. When one of those did a large piece of work, it queued up on the very component that everyone else's page was waiting on.

The symptom was never "an import failed". It was the portal going quiet while one was running, and, in the worst measured case, a burst of node writes holding up ordinary page subscriptions long enough for the portal to be restarted underneath its users.

What changes

Bulk imports, copies and moves now run on the portal's dedicated work queue, which exists for exactly this and has always been where the day-to-day writes go. Nothing about what they do changes: the same permissions are checked, by the same rules, as the same person.

For everything that was already off the switchboard, this changes nothing at all — which is the useful property. The hop is written so that it is a no-op unless the caller really was standing on the switchboard, so adopting it is never a behaviour change and never a judgement call.

And a new check keeps it that way. The same finding had been reported four separate times over the past month, each time from a different symptom, because nothing stopped it being reintroduced. The check names the file and the one-line fix at the moment the code is written, instead of a production log naming it weeks later.

What you will notice

• The portal stays responsive while a large import or copy is running.
• A guest session, and the first page after a restart, stop competing with routing for the few writes they need to do.

For the record

One piece of dead code was removed in the same pass: a second, unused copy of the node create/delete path that had never been wired to anything and did not take the hop. It could not misbehave while nothing called it, and it is the shape that would have been reintroduced by the first person who did.