policy-bot
is a GitHub App for enforcing
approval policies on pull requests. It does this by creating a status check,
which can be configured as a required status check.
While GitHub natively supports required reviews, policy-bot
provides more
complex approval features:
- Require reviews from specific users, organizations, or teams
- Apply rules based on the files, authors, or branches involved in a pull request
- Combine multiple approval rules with
and
andor
conditions - Automatically approve pull requests that meet specific conditions
Behavior is configured by a file in each repository. policy-bot
also provides a
UI to view the detailed approval status of any pull request.
- Designing Policies
- Configuration
- Security
- Deployment
- Development
- Contributing
- License
An approval policy is a set of rules that combine to determine when a pull request is approved. Policy Bot reports approval as a successful status check on the head commit of the pull request. The most common form of approval is code review from other developers, but Policy Bot supports many other types of approval.
Rules combine using the logical operators and
and or
. With no operators, a
policy requires that all of its rules are approved to approve the pull request.
Rules define the specific circumstances in which a pull request is approved. Each rule has four components:
-
Name. This is an arbitrary string that you use to refer to the rule in a policy. It also appears in the details view for a pull request. Try to pick names that make sense to humans and explain the purpose of the rule. A good name can complete the sentence "This pull request is approved if ..."
-
Predicates. A rule is only active when all of its predicates are true. If any predicate is false, the rule is skipped and does not influence the policy. Predicates allow you to create rules that are conditional on the state of the pull request, like requiring approval from a special team when specific important files change.
-
Options. Options control the behavior of the rule. This includes things like the approval methods, if pushing new changes invalidates previous approvals, and if the rule should request reviews from users or teams.
-
Requirements. These are the things that must be true for the rule to be approved. Requirements can be approvals from users with the right properties or can be conditions about the pull request, similar to predicates. If a rule has no requirements, it is always approved.
Only a name is required, but most rules need to include either predicates or requirements to be useful.
After evaluation, a rule can be in one of four states:
approved
- all of the predicates and requirements are truepending
- all of the predicates are true but one or more requirements are not trueskipped
- one or more predicates are not trueerror
- something went wrong while evaluating the rule
For the purposes of the and
and or
operators:
approved
is equivalent totrue
pending
anderror
are equivalent tofalse
skipped
completely removes the rule from the condition
When writing a rule, you can react to the state of pull request by using either predicates or required conditions. When would you use one over the other?
-
Use predicates to enable a rule when a pull request is in a specific state (or to skip the rule when a pull request is not in that state.) Predicates are useful when you want special approval requirements like requiring extra approval for files that involve security or automatically approving dependency updates.
-
Use required conditions when the pull request state itself is important for approval. For example, conditions are useful if you want to require that a pull request has certain status checks or specific labels.
Sometimes you can achieve a desired outcome using either approach. In this case, we prefer predicates. Policies that use predicates may define more rules, but tend to be flatter and use fewer logical operators. With well-chosen rule names, we find this style of policy easier to read and reason about.
By default, policies are defined in a .policy.yml
file at the root of the
repository. You can change this path and file name when running your own
instance of the server.
-
The file is read from the most recent commit on the target branch of each pull request.
-
The file may contain a reference to a policy in a different repository (see Remote Policy Configuration.)
-
If the file does not exist in the repository,
policy-bot
tries to load a sharedpolicy.yml
file at the root of the.github
repository in the same organization. You can change this path and repository name when running your own instance of the server, or disable this feature by settingoptions.shared_repository
to an empty string (""
) in the server configuration. -
If a policy does not exist in the repository or in the shared organization repository,
policy-bot
does not post a status check on the pull request. This means it is safe to enablepolicy-bot
on all repositories in an organization.
The overall policy is expressed by:
- Lists of rule definitions
- A set of policies that combine the rules or define additional options
Consider the following example, which allows changes to certain paths without
review, but all other changes require review from the palantir/devtools
team.
Any member of the palantir
organization can also disapprove changes.
# the high level policy
policy:
approval:
- or:
- the devtools team has approved
- only staging files have changed
disapproval:
requires:
organizations:
- "palantir"
# the list of rules
approval_rules:
- name: the devtools team has approved
requires:
count: 1
teams:
- "palantir/devtools"
- name: only staging files have changed
if:
only_changed_files:
paths:
- "^staging/.*$"
requires:
count: 0
The YAML language specification supports flow scalars (basic values like strings
and numbers) in three formats:
single-quoted,
double-quoted, and
plain. Each support different
escape characters, which can cause confusion when used for regex strings
(which often contain the \\
character).
- Single Quoted:
'
is used as an escape character. Backslash characters do not need to be escaped. e.g.'^BREAKING CHANGE: (\w| )+$'
- Double Quoted:
\
is used as an escape character. Backslash characters must be escaped with a preceding\
. e.g."^BREAKING CHANGE: (\\w| )+$"
- Plain: There are no escape characters. Backslash characters do not need to be escaped.
e.g.
^BREAKING CHANGE: (\w| )+$
You can also define a remote policy by specifying a repository, path, and ref
(only repository is required). Instead of defining a policy
key, you would
define a remote
key. Only 1 level of remote configuration is supported by design.
# The remote repository to read the policy file from. This is required, and must
# be in the form of "org/repo-name". The policy bot github app must have read
# access to this repository.
remote: org/repo-name
# The path to the policy config file in the remote repository. If none is
# specified, the default path in the server config is used.
path: path/to/policy.yml
# The branch (or tag, or commit hash) that should be used on the remote
# repository. If none is specified, the default branch of the repository is used.
ref: master
Each list entry in approval_rules
has the following specification:
# "name" is required, and is used to reference rules in the "policy" block
name: "example rule"
# "description" is optional and provides an explanation of the rule or
# additional help for users. Unlike YAML comments, it appears in the pull
# request details UI along with other information about the rule.
description: "A rule that explains how to configure all of the features"
# "if" specifies a set of predicates that must be true for the rule to apply.
# This block, and every condition within it are optional. If the block does not
# exist, the rule applies to every pull request.
if:
# "changed_files" is satisfied if any file in the pull request matches any
# regular expression in the "paths" list. If the "ignore" list is present,
# files in the pull request matching these regular expressions are ignored
# by this rule.
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
changed_files:
paths:
- "^config/.*$"
- "^server/views/.*\\.tmpl$"
ignore:
- "^config/special\\.file$"
# "no_changed_files" is the negation of "changed_files". This predicate is
# satisfied if no file in the pull request matches any regular expression
# in the "paths" list. If the "ignore" list is present, files in the pull
# request matching these regular expressions are ignored by this rule.
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
no_changed_files:
paths:
- "^other-config/.*$"
ignore:
- "^other-config/special\\.file$"
# "only_changed_files" is satisfied if all files changed by the pull request
# match at least one regular expression in the list.
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
only_changed_files:
paths:
- "^config/.*$"
# "has_author_in" is satisfied if the user who opened the pull request is in
# the users list or belongs to any of the listed organizations or teams. The
# `users` field can contain a GitHub App by appending `[bot]` to the end of
# the name, for example: `fun-github-app[bot]`
has_author_in:
users: ["user1", "user2", ...]
organizations: ["org1", "org2", ...]
teams: ["org1/team1", "org2/team2", ...]
# "has_contributor_in" is satisfied if any commits on the pull request have
# an author or committer in the users list or that belong to any of the
# listed organizations or teams.
has_contributor_in:
users: ["user1", "user2", ...]
organizations: ["org1", "org2", ...]
teams: ["org1/team1", "org2/team2", ...]
# "only_has_contributors_in" is satisfied if all of the commits on the pull
# request have an author or committer in the users list or that belong to
# any of the listed organizations or teams.
only_has_contributors_in:
users: ["user1", "user2", ...]
organizations: ["org1", "org2", ...]
teams: ["org1/team1", "org2/team2", ...]
# "author_is_only_contributor", when true, is satisfied if all commits in the
# pull request are authored by and committed by the user who opened the pull
# request. When false, it is satisfied if at least one commit in the pull
# request was authored or committed by another user.
author_is_only_contributor: true
# "targets_branch" is satisfied if the target branch of the pull request
# matches the regular expression
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
targets_branch:
pattern: "^(master|regexPattern)$"
# "from_branch" is satisfied if the source branch of the pull request
# matches the regular expression. Note that source branches from forks will
# have the pattern "repo_owner:branch_name"
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
from_branch:
pattern: "^(master|regexPattern)$"
# "modified_lines" is satisfied if the number of lines added or deleted by
# the pull request matches any of the listed conditions. Each expression is
# an operator (one of '<', '>' or '='), an optional space, and a number.
modified_lines:
additions: "> 100"
deletions: "> 100"
total: "> 200"
# DEPRECATED: Use "has_status" below instead, which is more flexible.
# "has_successful_status" is satisfied if the status checks that are specified
# are marked successful on the head commit of the pull request.
has_successful_status:
- "status-name-1"
- "status-name-2"
- "status-name-3"
# "has_status" is satisfied if the status checks that are specified are
# finished and concluded with one of the conclusions specified.
# "conclusions" is optional and defaults to ["success"].
has_status:
conclusions: ["success", "skipped"]
statuses:
- "status-name-1"
- "status-name-2"
- "status-name-3"
# "has_workflow_result" is satisfied if the GitHub Actions workflow runs that
# are specified all finished and concluded with one of the conclusions
# specified. "conclusions" is optional and defaults to ["success"].
# `workflows` contains the paths to the workflow files that are being checked.
# If a workflow is run more than once for a commit - for example for a `push`
# and `pull_request` event, the most recent completed run for each event type
# will be considered.
has_workflow_result:
conclusions: ["success", "skipped"]
workflows:
- ".github/workflows/a.yml"
- ".github/workflows/b.yml"
# "has_labels" is satisfied if the pull request has the specified labels
# applied
has_labels:
- "label-1"
- "label-2"
# "repository" is satisfied if the pull request repository matches any one of the
# patterns within the "matches" list or does not match all of the patterns
# within the "not_matches" list.
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
repository:
matches:
- "^palantir/policy.*$"
not_matches:
- "^palantir/.*docs$"
# "title" is satisfied if the pull request title matches any one of the
# patterns within the "matches" list or does not match all of the patterns
# within the "not_matches" list.
# e.g. this predicate triggers for titles including "BREAKING CHANGE" or titles
# that are not marked as docs/style/chore changes (using conventional commits
# formatting)
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
title:
matches:
- "^BREAKING CHANGE: (\\w| )+$"
not_matches:
- "^(docs|style|chore): (\\w| )+$"
# "has_valid_signatures" is satisfied if the commits in the pull request
# all have git commit signatures that have been verified by GitHub
has_valid_signatures: true
# "has_valid_signatures_by" is satisfied if the commits in the pull request
# all have git commit signatures that have been verified by GitHub, and
# the authenticated signatures are attributed to a user in the users list
# or belong to a user in any of the listed organizations or teams.
has_valid_signatures_by:
users: ["user1", "user2", ...]
organizations: ["org1", "org2", ...]
teams: ["org1/team1", "org2/team2", ...]
# "has_valid_signatures_by_keys" is satisfied if the commits in the pull request
# all have git commit signatures that have been verified by GitHub, and
# the authenticated signatures are attributed to a GPG key with an ID in the list.
has_valid_signatures_by_keys:
key_ids: ["3AA5C34371567BD2"]
# "options" specifies a set of restrictions on approvals. If the block does not
# exist, the default values are used.
options:
# If true, approvals by the author of a pull request are considered when
# calculating the status. False by default.
allow_author: false
# If true, the approvals of someone who has committed to the pull request are
# considered when calculating the status. The pull request author is considered
# a contributor. If allow_author and allow_contributor would disagree, this option
# always wins. False by default.
allow_contributor: false
# If true, the approvals of someone who has committed to the pull request are
# considered when calculating the status. In this case, pull request author is NOT
# considered a contributor. If combined with any combination of allow_author: true
# or allow_contributors: true, then the pull request author IS considered when
# calculating approval. False by default.
allow_non_author_contributor: false
# If true, pushing new commits to a pull request will invalidate existing
# approvals for this rule. False by default.
invalidate_on_push: false
# If true, comments on PRs, the PR Body, and review comments that have been edited in any way
# will be ignored when evaluating approval rules. Default is false.
ignore_edited_comments: false
# If true, "update merges" do not invalidate approval (if invalidate_on_push
# is enabled) and their authors/committers do not count as contributors. An
# "update merge" is a merge commit that was created in the UI or via the API
# and merges the target branch into the pull request branch. These are
# commonly created by using the "Update branch" button in the UI.
ignore_update_merges: false
# If present, commits authored and committed by users meeting the conditions
# are ignored for the purposes of approval. This means the users will not
# count as contributors and their commits will not invalidate approval if
# invalidate_on_push is enabled. Both the author and the committer must match
# the conditions to ignore the commit. This option has security implications,
# see the README for more details.
ignore_commits_by:
users: ["bulldozer[bot]"]
organizations: ["org1"]
teams: ["org1/team1"]
# Automatically request reviewers when a Pull Request is opened
# if this rule is pending, there are no assigned reviewers, and if the
# Pull Request is not in Draft.
# Reviewers are selected based on the set of requirements for this rule
# and reviewers can be augmented using the mode option.
request_review:
# False by default
enabled: true
# mode modifies how reviewers are selected. `all-users` will request all users
# who are able to approve the pending rule. `random-users` selects a small
# set of random users based on the required count of approvals. `teams` will
# request teams to review. Teams must have explicit access defined under
# https://github.com/<org>/<repo>/settings/access in order to be tagged,
# at least until https://github.com/palantir/policy-bot/issues/165 is fixed.
# Defaults to 'random-users'.
mode: all-users|random-users|teams
# count sets the number of users requested to review the pull request when
# using the `random-users` mode. If count is not set or set to 0, request the
# number of users set by requires.count. Setting this is useful when you want
# to request more reviewers than the required count. Defaults to 0.
count: 0
# "methods" defines how users may express approval.
methods:
# If a comment contains a string in this list, it counts as approval. Use
# the "comment_patterns" option if you want to match full comments. The
# default values are shown.
comments:
- ":+1:"
- "👍"
# If a comment matches a regular expression in this list, it counts as
# approval. Defaults to an empty list.
#
# Note: Double-quote strings must escape backslashes while single/plain do not.
# See the Notes on YAML Syntax section of this README for more information.
comment_patterns:
- "^Signed-off by \\s+$"
# If true, GitHub reviews can be used for approval. All GitHub review approvals
# will be accepted as approval candidates. Default is true.
github_review: true
# Just like the "comment_patterns" option, but for GitHub reviews. Only GitHub
# review approvals matching the included patterns will be accepted as
# approval candidates. Defaults to an empty list.
github_review_comment_patterns:
- '\b(?i)domain\s*lgtm\b'
# Just like the "comment_patterns" and "github_review_comment_patterns" option, but
# for the PR Body description. If a PR body contains a string in this list, it counts as approval. Use
# the "body_patterns" option if you want to match strings.
body_patterns:
- "\b(?i)no-platform"
# "requires" specifies the approval requirements for the rule. If the block
# does not exist, the rule is automatically approved.
requires:
# "count" is the number of required approvals. The default is 0, meaning no
# approval is necessary.
count: 1
# A user must be in the list of users or belong to at least one of the given
# organizations or teams for their approval to count for this rule.
users: ["user1", "user2"]
organizations: ["org1", "org2"]
teams: ["org1/team1", "org2/team2"]
# A user must have at least the minimum permission in this list for their
# approval to count for this rule. Valid permissions are "admin", "maintain",
# "write", "triage", and "read".
#
# Specifying more than one permission is only useful to control which users
# or teams are selected for review requests. See the documentation on review
# requests for details.
permissions: ["write"]
# "conditions" is the set of conditions that must be true for the rule to
# count as approved. If present, conditions are an additional requirement
# beyond the approvals required by "count".
#
# For example, if "count" is 1 and "conditions" contains the "has_status"
# condition with the "build" status, the rule is only approved once the
# "build" status check passes and one authorized reviewer leaves a review.
conditions:
# The "conditions" block accepts all of the keys documented as part of the
# "if" block for predicates. The "has_status" key is shown here as an
# example of one type of condition.
has_status:
statuses:
- "build"
- "vulnerability scan"
The approval
block in the policy
section defines a list of rules that must
all be true:
policy:
approval:
- rule1
- rule2
- rule3
- ...
Each list entry may be the name of a rule, or one of the following conjunctions:
or:
- rule1
- rule2
- ...
and:
- rule1
- rule2
- ...
Conjunctions can contain more conjunctions (up to a maximum depth of 5):
- or:
- rule1
- rule2
- and:
- rule3
- rule4
Disapproval allows users to explicitly block pull requests if certain changes must be made. Any member of in the set of allowed users can disapprove a change or revoke another user's disapproval.
Unlike approval, all disapproval predicates and options are specified as part
of the policy. Effectively, there is a single disapproval rule. The disapproval
policy has the following specification:
# "disapproval" is the top-level key in the policy block.
disapproval:
# "if" specifies a set of predicates which will cause disapproval if any are
# true
#
# This block, and every condition within it are optional. If the block does
# not exist, a pull request is only disapproved if a user takes a disapproval
# action.
if:
# All predicates from the approval rules section are valid here
title:
not_matches:
- "^(fix|feat|chore): (\\w| )+$"
- "^BREAKING CHANGE: (\\w| )+$"
matches:
- "^BLOCKED"
# "options" sets behavior related to disapproval. If it does not exist, the
# defaults shown below are used.
options:
# "methods" defines how users set and revoke disapproval.
methods:
# "disapprove" sets the methods for disapproval.
disapprove:
comments:
- ":-1:"
- "👎"
github_review: true
# "revoke" sets the methods for revoking disapproval. Usually, these will
# match the methods used by approval rules.
revoke:
comments:
- ":+1:"
- "👍"
github_review: true
# "requires" sets the users that are allowed to disapprove. If it is not set,
# disapproval is not enabled.
requires:
users: ["user1", "user2"]
organizations: ["org1", "org2"]
teams: ["org1/team1", "org2/team2"]
Sometimes it is useful to test if a given policy file is valid, especially in a CI environment.
An API endpoint exists at /api/validate
to validate the syntax of the yaml and policy configuration,
however it cannot validate that the rules are semantically correct for a given use case.
The API can be used as such:
$ curl https://policybot.domain/api/validate -XPUT -T path/to/policy.yml
{"message":"failed to parse approval policy: failed to parse subpolicies for 'and': policy references undefined rule 'the devtools team has approved', allowed values: [the devtools team has]","version":"1.12.5"}
You can examine the HTTP response code to automatically detect failures
$ rcode=$(curl https://policybot.domain/api/validate -XPUT -T path/to/policy.yml -s -w "%{http_code}" -o /tmp/response)
$ if [[ "${rcode}" -gt 299 ]]; then cat /tmp/response && exit 1; fi
It can be useful to simulate how Policy Bot would evaluate a pull request if certain conditions were changed. For example: adding a review from a specific user or group, or adjusting the base branch.
An API endpoint exists at api/simulate/:org/:repo/:prNumber
to simiulate the result of a pull request. Simulations using this endpoint will NOT write the result back to the pull request status check and will instead return the result.
This API requires a GitHub token be passed as a bearer token. The token must have the ability to read the pull request the simulation is being run against.
The API can be used as such:
$ curl https://policybot.domain/api/simulate/:org/:repo/:number -H 'authorization: Bearer <token>' -H 'content-type: application/json' -X POST -d '<data>'
Currently the data payload can be configured with a few options:
Ignore any comments from specific users, team members, org members or with specific permissions
{
"ignore_comments":{
"users":["ignored-user"],
"teams":["ignored-team"],
"organizations":["ignored-org"],
"permissions":["admin"]
}
}
Ignore any reviews from specific users, team members, org members or with specific permissions
{
"ignore_reviews":{
"users":["ignored-user"],
"teams":["ignored-team"],
"organizations":["ignored-org"],
"permissions":["admin"]
}
}
Simulate the pull request as if the following comments from the following users had also been added
{
"add_comments":[
{
"author":"not-ignored-user",
"body":":+1:",
"created_at": "2020-11-30T14:20:28.000+07:00",
"last_edited_at": "2020-11-30T14:20:28.000+07:00"
}
]
}
Simulate the pull request as if the following reviews from the following users had also been added
{
"add_reviews":[
{
"author":"not-ignored-user",
"state": "approved",
"body": "test approved review",
"created_at": "2020-11-30T14:20:28.000+07:00",
"last_edited_at": "2020-11-30T14:20:28.000+07:00"
}
]
}
Choose a different base branch when simulating the pull request evaluation
{
"base_branch": "test-branch"
}
The above can be combined to form more complex simulations. If a Simulation is run without any data being passed, the pull request is evaluated as is.
There are several additional behaviors that follow from the rules above that are worth mentioning.
You must set at least one of the disapproval.requires
fields to enable
disapproval. Without setting one of these fields, GitHub reviews that request
changes have no effect on the policy-bot
status.
GitHub Reviews allow a user to dismiss the last review they left, causing it to
no longer count towards rule evaluations. When this happens policy-bot
will
use a previous, non-dismissed review, if it exists, when evaluating rules.
For example, if a user leaves an "approval" review and follows up with a
"request changes" review, policy-bot
will use the "request changes" review
when evaluating rules. However, if the user then dimisses their "request
changes" review, policy-bot
will instead use the initial "approval" review in
evaluating any rules.
If the if
block of a rule (the predicate) is not satisfied, the rule is
marked as "skipped". Skipped rules interact with or
and and
as follows:
- An
and
block containing only skipped rules is also skipped - An
or
block containing only skipped rules is also skipped
Effectively, skipped rules are treated as if they don't exist.
policy-bot
allows approval rules to reference organizations and teams that are
not in the organization that owns the repository where the rules appear. In
this case, policy-bot
must be installed on all referenced organizations.
For a commit on a branch to count as an "update merge" for the purpose of the
ignore_update_merges
option, the following must be true:
- The commit must have exactly two parents
- The commit must have the
committedViaWeb
property set totrue
- The first parent must exist in the pull request while the second parent must not exist in the pull request (meaning it is on the target branch)
These will all be true after updating a branch using the UI, but historic merges on long-running branches or merges created with the API may not be ignored. If this happens, you will need to reapprove the pull request.
This feature has security implications.
policy-bot
can automatically request reviewers for all pending rules
when Pull Requests are opened by setting the request_review
option.
The mode
enum modifies how reviewers are selected. There are currently three
supported options:
all-users
to request all users who can approverandom-users
to randomly select the number of users that are requiredteams
to request teams for review. Teams must be repository collaborators with at least read access.
options:
request_review:
enabled: true
mode: all-users|random-users|teams
The set of requested reviewers will not include the author of the pull request or users who are not collaborators on the repository.
When requesting reviews for rules that use repository permissions to select
approvers, only users who are direct collaborators or members of
repository teams are eligible for review selection. The users or their teams
must be granted an exact permission specified in the permissions
list of the
rule.
For example, if a rule can be approved by any user with admin
permission,
only direct or team admins are selected for review. Users who inherit
repository admin
permissions as organization owners are not selected.
The teams
mode needs the team visibility to be set to visible
to enable this functionality for a given team.
Given the following example requirement rule,
requires:
count: 2
users: ["user1", "user2"]
organizations: ["org1", "org2"]
teams: ["org1/team1", "org2/team2"]
policy-bot
will attempt to request 2 reviewers randomly from the expanded
set of users of in
["user1", "user2", "users in org1", "users in org2", "users in org1/team1", "users in org2/team"]
Where the Pull Request Author and any non direct collaborators have been removed from the set.
By default, policy-bot
does not invalidate exisitng approvals when users add
new commits to a pull request. You can control this behavior for each rule in a
policy using the invalidate_on_push
option.
To invalidate approvals, policy-bot
compares an estimate of the push time of
each commit with the time of each approval comment or review. The push time
estimate uses the time of the oldest status check, or the current time during
evaluation if there are no status checks. This is guaranteed to be after the
actual push time, but the delay may be arbitrarily large based on GitHub
webhook delivery behavior and processing time in policy-bot
.
In practice, this means that adding an approval immediately after (within a few
seconds of) a push may not approve the pull request. If this happens, leave a
second approval comment or review after policy-bot
adds the "pending" status
check.
policy-bot
caches push times in memory to improve performance and reduce API
requests.
Older versions of policy-bot
(before 1.31.0) used the pushedDate
field in
GitHub's GraphQL API to estimate commit push times. GitHub removed this field
in mid-2023 because computing it was unreliable and inaccurate (see issue
#598 for more details.)
The details view for a pull request shows the users, organizations, teams, and
permission levels that are reqired to approve each rule. When the
options.expand_required_reviewers
server option is set, policy-bot
expands
these to show the list of users whose approval will satisfy each rule. This can
make it easier for developers to figure out who they should ask for approval.
Like with review requests, when expanding permission levels only users with collaborator permissions on the repository, either directly or via teams, are included in the expanded list.
Enabling this option can expose otherwise private information about teams, organizations, and permissions to any user with read permission on a pull request. This includes teams in organizations other than the one that contains the pull request.
As a result, only enable this feature if all users with access to policy-bot
are allowed to view the members and permissions of any organization that uses
policy-bot
.
While policy-bot
can be used to implement security controls on GitHub
repositories, there are important limitations to be aware of before adopting
this approach.
policy-bot
reports approval status to GitHub using commit statuses. While
statuses cannot be deleted, they can be set or overwritten by any user with
write access to a repository. To prevent forged statuses, GitHub allows setting
an expected source for a status check when making it a requirement on a
protected branch. Policy Bot always should be set as the expect source for
its checks.
For older versions of GitHub Enterprise that do not support expected sources
for status checks, policy-bot
contains an auditing feature to detect
overwritten statuses. In addition to logging an audit event, it will replace
the forged status with a failure. However, a well-timed attempt can still
approve and merge a pull request before policy-bot
can detect the problem.
Organizations concerned about this case should monitor and alert on the
relevant audit logs or minimize write access to repositories.
GitHub users with sufficient permissions can edit the comments of other users,
possibly changing an unrelated comment into one that enables approval.
policy-bot
also contains audting for this event, but as with statuses, a
well-timed edit can approve and merge a pull request before policy-bot
can
detect the problem. Organizations concerned about this case can use the
ignore_edited_comments
option or can monitor and alert on the relevant audit
logs.
This issue can also be minimized by only using GitHub reviews for approval, at the expense of removing the ability to self-approve pull requests.
GitHub associates commits with users by mapping the email address in a commit
to email addresses associated with GitHub user accounts. policy-bot
then uses
the GitHub username to evaluate user-based rules and options. There are two
failure modes in this process:
-
If GitHub does not recognize either the author or committer email of a commit,
policy-bot
cannot evaluate the commit with respect to user-based rules and the commit is effectively ignored. -
If emails are manipulated when creating a commit, a user can trick GitHub and
policy-bot
into attributing the commit to a different user.
If using GitHub Enterprise, both of these issues are avoidable by using the commit-current-user-check pre-receive hook.
When using the ignore_update_merges
option, policy-bot
cannot tell the
difference between clean merges and merges that contain conflict resolution.
This means that a user who carefully crafts a pull request to generate a
conflict can use the web conflict editor to add unapproved changes to the file
containing the conflict.
Depending on the author of the merge commits, it may be possible to avoid this
issue by using the ignore_commits_by
option in combination with the
commit-current-user-check pre-receive hook.
policy-bot
is easy to deploy in your own environment as it has no dependencies
other than GitHub. It is also safe to run multiple instances of the server,
making it a good fit for container schedulers like Nomad or Kubernetes.
We provide both a Docker container and a binary distribution of the server:
- Binaries: https://github.com/palantir/policy-bot/releases
- Docker Images: https://hub.docker.com/r/palantirtechnologies/policy-bot/
A sample configuration file is provided at config/policy-bot.example.yml
.
Certain values may also be set by environment variables; these are noted in the
comments in the sample configuration file. By default, the environment
variables for server values are prefixed with POLICYBOT_
(e.g.
POLICYBOT_PORT
). This prefix can be overridden by setting the
POLICYBOT_ENV_PREFIX
environment variable.
To configure policy-bot
as a GitHub App, set these options in GitHub:
- Under Identifying and authorizing users
- Set User authorization callback URL to
http(s)://<your-policy-bot-domain>/api/github/auth
- Uncheck Request user authorization (OAuth) during installation
- Set User authorization callback URL to
- Under Webhook
- Set Webhook URL to
http(s)://<your-policy-bot-domain>/api/github/hook
- Set Webhook secret: A random string that matches the value of the
github.app.webhook_secret
property in the server configuration
- Set Webhook URL to
The app requires these permissions:
Permission | Access | Reason |
---|---|---|
Actions | Read-only | Read workflow run events for the has_workflow_result predicate |
Repository contents | Read-only | Read configuration and commit metadata |
Checks | Read-only | Read check run results |
Repository administration | Read-only | Read admin team(s) membership |
Issues | Read-only | Read pull request comments |
Merge Queues | Read-only | Read repository merge queues |
Repository metadata | Read-only | Basic repository data |
Pull requests | Read & write | Receive pull request events, read metadata. Assign reviewers |
Commit status | Read & write | Post commit statuses |
Organization members | Read-only | Determine organization and team membership |
The app should be subscribed to these events:
- Check run
- Issue comment
- Merge groups
- Pull request
- Pull request review
- Status
- Workflow Run
There is a logo.png
provided if you'd like to use it as the GitHub application logo. The background
color is #4d4d4d
.
After creating the app, update the server configuration file with the following generated values:
- App ID (
github.app.integration_id
) - Client ID (
github.oauth.client_id
) - Client secret (
github.oauth.client_secret
) - Private key (
github.app.private_key
)
policy-bot
uses go-baseapp and
go-githubapp, both of which emit
standard metrics and structured log keys. Please see those projects for
details.
To develop policy-bot
, you will need a Go installation.
If you want to build the UI, you'll also need NodeJS
and Yarn.
Run style checks and tests
./godelw verify
Running the server locally
# copy and edit the server config
cp config/policy-bot.example.yml config/policy-bot.yml
./godelw run policy-bot server
config/policy-bot.yml
is used as the default configuration file- The server is available at
http://localhost:8080/
Installing UI dependencies and building assets
# install dependencies
yarn install
# build CSS and JS assets
yarn run build
-
This generates a combined stylesheet with
policy-bot
styles and Tailwind core styles. It also copies JS files and other assets into the correct locations. -
To use the local asset files with a local server, add or uncomment the following in the server configuration file:
files: static: build/static templates: server/templates
Running the server via docker
# copy and edit the server config
cp config/policy-bot.example.yml config/policy-bot.yml
# build the docker image
./godelw docker build --verbose
docker run --rm -v "$(pwd)/config:/secrets/" -p 8080:8080 palantirtechnologies/policy-bot:latest
- This will mount the path relative path
config/
which should contain the modified config filepolicy-bot.yml
- The server is available at
http://localhost:8080/
Example policy files can be found in config/policy-examples
Contributions and issues are welcome. For new features or large contributions, we prefer discussing the proposed change on a GitHub issue prior to a PR.
This library is made available under the Apache 2.0 License.