Thread.Current storage under unicorn/thin

[toxaq]

When using thin or unicorn Thread.Current isn't guaranteed to be cleared between requests, so you could end up having a tenant set when you don't want it set. As per this SO Question

There seems to be two solutions here:

1. Add a note to the readme to create a before_filter that sets the current_tenant to nil if you aren't explicitly setting a tenant for each request
2. Use something like RequestStore as referenced in this rails issue thread

Thoughts?

[jfrux]

Is this related to race conditions?
Could it be possible for a record to get saved to an incorrect tenant if the user requests a page on a different tenant right before the save takes place...?

[toxaq]

Only if you aren't guaranteed to set a tenant in each request. In my case I had two namespaced sections of the site, one which guaranteed a tenant was set and one which didn't. The section not setting a tenant would often throw errors because a tenant was still set when it shouldn't have been (and hadn't been during that request). I now explicitly set the tenant to nil on the non-tenant section.

[ErwinM]

I read the SO post. Interesting stuff.

I'm thinking of clearing the Thread.current[:current_tenant] explicitly, but I'm having trouble seeing all the use cases.

@toxaq In the app you are talking about in your comment: on the section without the tenant set, is the code still passing through the tenant assignment in application_controller (set_tenant_by_subdomain or using the before_filter approach) or is it somehow skipping it?

[toxaq]

Yeah, I'm using the before filter approach. I only set it if required, so it effectively skipped the setting. Setting to nil explicitly works fine (my fix).

[ErwinM]

I've switched the gem (master) to using request_store instead of Thread.current. This should ensure a clean slate for each request.

@toxaq If you could find the time to test if this indeed fixes the issues you were experiencing (without your later fixes), that would be much appreciated!

[toxaq]

Just caught up on this, will try a test this week and see how I go. Thanks for the update!

[swordfish444]

@toxaq Were you ever able to determine the issue and what resolved it?

[toxaq]

@swordfish444 it's a long time ago now but I think Erwin's change to request_store resolved this but explicitly setting on every request in a before_filter also worked adequately.