patch-manager

This tool is used to automate OpenShift z-stream patch manager daily routine by listing all z-stream candidate pull requests and triaging them based on multiple criteria.

Triage is done via scoring system where each pull request is being classified by the following:

Bug Severity
Bug Keywords
PM Score
Component Priority

Complete implementation and details can be found in pkg/classifiers package.

Each classifier can be configured via config.yaml file. Most of them assign "score" to each pull request. The score is additive, so in the example below, an "urgent" bug with TestBlocker keyword will get score 1.8 which will likely put it at the top of the list.

Example release config YAML:

---
# MergeWindow describe a time window when pull requests can be cherry-picked for the z-stream.
mergeWindow:
  from: # YYYY-MM-DD
  to: # YYYY-MM-DD
# Capacity describe the QE capacity for the "next" week per QE group.
capacity:
  default: 5 # <- this is a "default" capacity if no capacity is specified for a component
  groups:
    - name: API&Auth
      capacity: 5 # <- this is the QE capacity for all components listed below
      components:
      - Master
      - apiserver-auth
      - authentication
      - service-ca
      - openshift-apiserver
      - oauth-apiserver
      - kube-apiserver
      - oauth-proxy
      - kube-storage-version-migrator
      - config-operator
    - name: Workloads
      capacity: 5
      components:
      - Deployments
      - Command Line Interface
      - oc
      - kube-controller-manager
      - kube-scheduler
# Rules specify go/no-go rules to apply on every PR. These rules does not use classification/scoring, but make decisions directly
# based on conditions
rules:
  labels:
    # Refuse lists pull request label prefixes that when found the pull request will be automatically "skipped" (reason will be recorded)
    refuse:
      - do-not-merge/hold
# Classifiers describe how much score points a single pull request should get. (0-1)
# Score impact the position of a PR in merge queue.
classifiers:
  # Keywords classifier assign score based on bugzilla keywords present in bug associated with pull request
  keywords:
    "TestBlocker": 0.8
    "UpgradeBlocker": 0.8
    "Security": 0.5
  # Components classifier assign score based on importance/criticality of components
  components:
    "authentication": 0.5
    "networking": 0.5
    "node": 0.5
    "kube-apiserver": 0.5
  # Severities classifier assign score based on bug Severity field
  severities:
    "urgent": 1.0
    "high": 0.5
    "medium": 0.2
    "low": 0.1
    "unknown": -1.0
  # PMScores classifier assign score based on PMScore field ranges
  pmScores:
    - from: 0
      to: 30
      score: 0
    - from: 30
      to: 50
      score: 0.2
    - from: 50
      to: 100
      score: 0.5
    - from: 100
      to: 999
      score: 0.8

Using the above config, only 2 pull requests for Networking component will be approved and only 5 pull requests for kube-apiserver bugzilla component will be picked.

Pre-requirements

You will need Github Personal Token exported via environment variable GITHUB_TOKEN (or use command flag).
You will need Bugzilla API Token exported via environment variable BUGZILLA_APIKEY (or use command flag).
You need a config file (check release/ directory). You can also use PATCHMANAGER_CONFIG=https://raw.githubusercontent.com/openshift/patchmanager/main/release/4.7.yaml

Installation

To install this command run:

$ go get github.com/openshift/patchmanager/cmd/patchmanager

The binary patchmanager should be installed in your $GOPATH/bin directory.

Alternatively you can use podman to run patchmanager. Example:

podman run -it --env-host quay.io/mfojtik/patchmanager:latest patchmanager run

Development

In case you want to contribute, you can use Makefile to build the patchmanager binary by invoking make.

Usage

Run patchmanager run --config=path/to/config.yaml -o candidates.yaml will produce YAML file of candidate pull request for 4.x release already sorted and scored based on the classifiers. The capacity flag will cause that on N pull requests will be "picked". (TIP: You can set the PATCHMANAGER_CONFIG environment variable which points to a config location)

Example YAML file:

items:
- pullRequest:
    # Description: MCDDrainError "Drain failed on  , updates may be blocked" missing
    # rendered node name
    # Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1906298
    # Component: Machine Config Operator
    # Severity: high
    # PM Score 52
    url: https://github.com/openshift/machine-config-operator/pull/2419
    decision: pick
    score: 1.0
- pullRequest:
    # Description: 4.7 to 4.6 downgrade fails due to 4.7 Cluster Profile Support manifest
    # changes
    # Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1925199
    # Component: Cluster Version Operator
    # Severity: medium
    # PM Score 77
    url: https://github.com/openshift/cluster-version-operator/pull/512
    decision: pick
    score: 0.7
- pullRequest:
    # Description: [Kuryr] Available port count not correctly calculated for alerts
    # Bug: https://bugzilla.redhat.com/show_bug.cgi?id=1897526
    # Component: Networking
    # Severity: low
    # PM Score 0
    url: https://github.com/openshift/cluster-network-operator/pull/907
    decision: skip
    score: 0.6
    decisionReason: target capacity for component Networking is 2

A human patch manager need to review this YAML file and make decisions on individual changes. Decision can be either pick or skip. The decisionReason field will be used in a comment if -add-comment flag is specified (see below).
Once you are done editing YAML file, you can run the patchmanager approve --config=path/to/config.yaml -f candidates.yaml command which will apply the cherry-pick-approved label on ALL pull requests with "pick" decision. If --skip-comment or/and --pick-comment are set, then a comment will be made to the PR about decision reason. If these are not used, not comment will be made on PR.
Alternatively, you can use patchmanager list -f candidates.yaml to format the pull requests in human readable table:

$ patchmanager list -f candidates.yaml 
  URL (35)                                                                              SCORE   DECISION   REASON                                                      
 ------------------------------------------------------------------------------------- ------- ---------- ------------------------------------------------------------ 
  https://github.com/openshift/ovn-kubernetes/pull/460                                   2.00   pick                                                                   
  https://github.com/openshift/machine-config-operator/pull/2462                         1.50   pick                                                                   
  https://github.com/openshift/machine-config-operator/pull/2426                         1.50   pick                                                                   
  https://github.com/openshift/openshift-apiserver/pull/187                              1.00   pick                                                                   
  https://github.com/openshift/cluster-ingress-operator/pull/570                         0.20   skip       maximum picks set by patch manager for this z-stream is 10  
  https://github.com/openshift/origin/pull/25913                                         0.20   skip       maximum picks set by patch manager for this z-stream is 10  
  https://github.com/operator-framework/operator-lifecycle-manager/pull/2036             0.20   skip       maximum picks set by patch manager for this z-stream is 10  
  https://github.com/openshift/console-operator/pull/512                                 0.20   skip       maximum picks set by patch manager for this z-stream is 10

mrunalp / patchmanager