4.6 KiB
Design Decisions
This file documents the design choices as well as configuration options.
Stateless
No state storage is needed on renovate or the source code repository apart
from what you see publicly (branches, Pull Requests). It, therefore, doesn't
matter if you stop/restart the script and would even still work if you had it
running from two different locations, as long as their configuration was the
same.
Synchronous Operation
The script current processes repositories and branches within them all synchronously.
- Greatly reduces the chance of hitting simultaneous API rate limits
- Simplifies logging
Cascading Configuration
Configuration options applied per-package (e.g. with package rules) override those applied per package-type, which override those per-repository, which override those which are global (all repositories).
Automatic discovery of package file locations
The default behaviour is to auto-discover all package file (e.g. package.json) locations in a repository and process them all. Doing so means that "monorepos" are supported by default. This can be overridden by the configuration option includePaths, where you list the file paths manually (e.g. limit to just package.json in the root of the repository).
Separate Branches per dependency
By default, renovate will maintain separate branches per-dependency. So if 20
dependencies need updating, there will be at least 20 branches/PRs. Although
this may seem undesirable, it was considered even less desirable if all 20 were
in the same Pull Request and it's up to the users to determine which dependency
upgrade(s) caused the build to fail.
However, it's still possible to override the default branch and PR name
templates in such a way to produce a single branch for all dependencies. The
groupName configuration option can be used at a repository level (e.g. give it
the value All) and then all dependency updates will be in the same branch/PR.
Separate Minor and Major PRs
renovate will create multiple branches/PRs if both major and minor branch upgrades
are available. For example, if the current example is 1.6.0 and upgrades to 1.7.0
and 2.0.0 exist, then renovate will raise PRs for both the 1.x upgrade(s) and
2.x upgrade(s).
- It's often the case that projects can't upgrade major dependency versions immediately.
- It's also often the case that previous major versions continue receiving Minor or Patch updates.
- Projects should get Minor and Patch updates for their current Major release even if a new Major release exists
This can be overridden via the config option separateMajorMinor.
Branch naming
Branches are named like renovate/webpack-1.x instead of renovate/webpack-1.2.0.
- Branches often receive updates (e.g. new patches) before they're merged.
- Naming the branch like
1.xmeans its name still names sense if a1.2.1release happens
Note: Branch names are configurable using string templates.
Pull Request Recreation
By default, the script does not create a new PR if it finds an identical one already closed. This allows users to close unwelcome upgrade PRs and not worry about them being recreated every run. Typically this is most useful for major upgrades. This option is configurable.
Rebasing Unmergeable Pull Requests
With the default behaviour of one branch per dependency, it's often that case that a PR gets merge conflicts after an adjacent dependency update is merged. Although platforms often have a web interface for simple merge conflicts, this is still annoying to resolve manually.
renovate will rebase any unmergeable branches and add the latest necessary
commit on top of the most recent master commit.
Note: renovate will only do this if the original branch hasn't been modified
by anyone else.
Suppressing string templates from CLI
String templates (e.g. commit or PR name) are not configurable via CLI options, in order to not pollute the CLI help and make it unreadable. If you must configure via CLI, use an environment variable instead. e.g.
$ RENOVATE_BRANCH_NAME=foo renovate
Alternatively, consider using a Configuration File.
Logging and error levels
Renovate uses the following convention for log levels:
- logger.error should only be used for problems that are likely to be a Renovate bug or require Renovate improvements. These are the types of errors that Renovate administrators should be alerted to immediately
- logger.warn should be used for problems that might be a Renovate problem so should be checked periodically in batches
- For user problems (e.g. configuration errors), these should not warn or error on the server side and instead use logger.info