test config before reloading [untested]

[schmunk42]

⚠️ This is an untested change for now ⚠️

Question/discussion:

In general, wouldn't it be better to test the config before reloading nginx.
The expected outcome would be to have nginx still running, with your old config - instead of loosing connection through reverse-proxy to all your sites.

Related to

• Unable to profixy a container started with docker-compose version 2  #502
• Error: no servers are inside upstream in #438

[schmunk42]

CC: @handcode

[thomasleveil]

even with this, if you restart the nginx-container, nginx will fail to start. This just delays the problem IMHO.

And if you create new containers (using VIRTUAL_HOST) those will be ignored. I'm afraid users will then wonder why their new container is not considered by the nginx-proxy and this will end up in a support request here.

[schmunk42]

@thomasleveil Yes that's correct.

But the issue was more like, if there's a broken config somehow and you try to reload, nginx will fail and all your virtual-hosts will be offline, i.e. it's related to #585

I'd say to original behavior might be more desirable in development, while the patch is definitely the way to go in production.

What's about using an ENV variable for this?

[schmunk42]

Any news here? Without a fix for this we end up loosing connection to all sites, just because of one broken config/setup.