As mentioned by @sdktr the "--install-option" is redundant. The added
"--no-warn-script-location" is used to silence warnings when installing
into a directory that is not on $PATH.
With the new multistage build the pip install command put some packages
in the wrong directory. This resuled in them not being copied into the
runtime images. This commit add the additional "--prefix" option to the
install command. After that all test run and we no longer get the error
"ModuleNotFoundError: No module named 'drf_yasg'".
This changes the build process even further. Instead
f using `wget` to fetch the current code, `git` is used.
This allows for faster switching between branches,
because only the differences between them have to be
fetched from the server.
But the main advantage is that the build cache can
finally be used as designed by Docker. Repetitive
builds are very fast now. This is also true between
branches and tags, as long as the `requirements.txt`
file doesn't change.
This commit introduces a huge change in the build process.
What changed:
- Dockerfile.ldap was integrated into Dockerfile as a seperate
[build stage][multistage-build].
- All the build scripts were refactored according to this.
- The `docker-compose.yml` file was adjusted likewise.
- The main build script, `/build.sh`, now always builds all
targets (formerly called variants).
- The minimal requirements for Docker and docker-compose
have increased.
- The build on hub.docker.com must be adjusted.
This change should also fix#156 permanently.
[multistage-build]: https://docs.docker.com/develop/develop-images/multistage-build/
After [a discussion][1] with rje6459 on the networktocode Slack
in the #netbox-docker channel we've come up with a small improvement
to those that regularly have to build netbox docker locally.
It fetches the "requirements.txt" from the desired branch before
downloading netbox itself.
[1]: https://networktocode.slack.com/archives/CD23LP8BC/p1558537080023200
One container should ideally have one responsibility [1]. Therefore I
implemented the netbox-worker to start in it's own container. This is
possible, because netbox and the worker communicate via redis anyway.
They still use the same image underneath, just the "command" they
execute while starting different.
Or in other words: I see no reason to introduce supervisord, when we
already have docker-compose which can take care of running multiple
processes.
Also, here's another benefit: Now it's possible to view the logs of the
webhook worker independently of the other netbox logs (and vice-versa).
Other changes in this commit:
* I don't see a reason to put a password for Redis in the docker-compose
setup, so I removed it.
* Slightly changed the nginx config, so that the nginx startup command
becomes simpler and any error should be visible in the docker log.
* Some housekeeping in the `Dockerfile`.
* Added some troubleshooting advice regarding webhooks to the README.
I'd like to thank Brady (@bdlamprecht [2]) here who did the harder
work of figuring out what's even required to have webhooks working. [3]
[1]
https://docs.docker.com/develop/develop-images/dockerfile_best-practices/#decouple-applications
[2] https://github.com/bdlamprecht
[3] https://github.com/ninech/netbox-docker/pull/90
With this the configuration is moved to /etc/netbox/config and the
default reports directory is set to /etc/netbox/reports. This enables
the user to mount reports from a config map or persistent volume in
OpenShift.
With these modifications the netbox containers can run on Openshift.
The configuration files can be imported from a config map to the
'/etc/netbox' directory and will be loaded from there.
Initializers are startup scripts for common tasks like creating custom
fields. These are problems many users of Netbox Docker potentially face
and are therefore worth sharing.
The present build does not include a font-family causing Graphviz to render text as blank squares, including the ttf ubuntu font-family means text is rendered correctly
