system-config/tools/owners.py
Jeremy Stanley bf8e2b9ad0 Limit PTL rolls to foundation members
Missing the original I920e1bc134d1283e2019f4389a5491e9b746cca0
implementation, also filter the per-team PTL electorates to only
include OpenStack Foundation Individual Members. The previous change
did this correctly for the full (Technical Committee) electorate but
failed to apply a similar conditional constraint to Project Team
Lead electoral rolls.

Change-Id: I03e90e5c512ee463cfcc49f147ff523806ae05c7
Story: #2001094
Task: #4877
2017-08-10 00:20:45 +00:00

702 lines
27 KiB
Python

#!/usr/bin/env python
# Copyright (c) 2016 OpenStack Foundation
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing,
# software distributed under the License is distributed on an "AS
# IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either
# express or implied. See the License for the specific language
# governing permissions and limitations under the License.
# Description: When run using OpenStack's Gerrit server, this builds
# YAML representations of aggregate change owner details and change
# counts for each governance project-team, as well as a combined set
# for all teams.
# Rationale: The OpenStack Technical Committee and Project Team Lead
# elections need electorate rolls taken from "Active Technical
# Contributors" to any repos under official project-teams over a
# particular timeframe. Similarly, the OpenStack Foundation gives
# summit registration discount codes to contributors meeting similar
# criteria. The Gerrit REST API provides access to all the data
# necessary to identify these individuals.
# Use: The results end up in files named for each
# official governance project-team (or "all") ending with a .yaml
# extension. At the time of writing, it takes approximately 30
# minutes to run on a well-connected machine with 70-80ms round-trip
# latency to review.openstack.org.
# An example for generating the March 2016 technical election rolls:
#
# $ virtualenv venv
# [...]
# $ ./venv/bin/pip install pyyaml requests
# [...]
# $ ./venv/bin/python tools/owners.py -a 2015-03-04 \
# -b 2016-03-04 -i 11131 -i 22816 -o owners \
# -r march-2016-elections
# MISSING: ansible-build-image
# MERGING DUPLICATE ACCOUNT: 8074 into 2467
# [...blah, blah, blah...wait for completion...]
#
# TODO(fungi): Add a pass which will correctly generate the
# stable_branch_maintenance.* files. In the meantime, to properly
# generate the SBM PTL electorate, run a second time with a
# different -o of sbm, adding the -n and -s options, and then copy
# the full electorate over like:
#
# $ ./venv/bin/python tools/owners.py -a 2015-03-04 \
# -b 2016-03-04 -i 11131 -i 22816 -o sbm \
# -r march-2016-elections -n -s 'branch:^stable/.*'
# [...wait for completion again...]
# $ cp sbm/_electorate.txt owners/stable_branch_maintenance.txt
# $ cp sbm/_all_owners.yaml owners/stable_branch_maintenance.yaml
#
# Once complete, make a compressed tarball of the owners directory
# and send it attached to a PGP/MIME signed message to the appointed
# election officials. The various *.txt files are lists of the
# preferred addresses of all valid voters for the various PTL
# elections (whose team names correspond to the file names),
# suitable for passing directly to CIVS. The similarly named *.yaml
# files are detailed structured data about the same sets of voters,
# for use in validating the address lists. The _electorate.txt file
# is the equivalent address list for the TC election voters, and its
# corresponding structured data is in _all_owners.yaml.
# You can also do interesting analysis on _all_owners.yaml, for
# example:
#
# $ ./venv/bin/python
# >>> import yaml
# >>>
# >>> o = yaml.load(open('owners/_all_owners.yaml'))
# >>> for c in range(5):
# ... print('Owners of at least %s changes: %s' % (
# ... c+1,
# ... len({k: v for k, v in o.iteritems() if v['count'] > c})))
# ...
# Owners of at least 1 changes: 3239
# Owners of at least 2 changes: 2352
# Owners of at least 3 changes: 1924
# Owners of at least 4 changes: 1682
# Owners of at least 5 changes: 1504
from __future__ import print_function
import argparse
import csv
import datetime
import json
import os
import sys
import requests
import yaml
try:
from string import maketrans
except ImportError: # Python3
maketrans = bytes.maketrans
def dumper(data, stream):
"""Convenience wrapper to consistently set YAML formatting"""
return yaml.safe_dump(data, allow_unicode=True, default_flow_style=False,
encoding='utf-8', stream=stream)
def normalize_email(email):
"""Lower-case the domain part of E-mail addresses to better spot
duplicate entries, since the domain part is case-insensitive
courtesy of DNS while the local part is not necessarily"""
local, domain = email.split('@')
domain = domain.lower()
return '%s@%s' % (local, domain)
def normalize_project(project):
"""Replace spaces and hyphens with underscores in project teams
and then lower-case them, for more convenient filenames"""
return project.translate(maketrans(' -', '__')).lower()
def date_merged(change, after=None, before=None):
"""Determine the date and time a specific change merged"""
date = change.get('submitted', None)
if not date:
# Something's terribly wrong with any changes matching this now
print(
'SKIPPING DATELESS MERGE: change %s for account %s'
% (change['_number'], change['owner']['_account_id']),
file=sys.stderr)
return None
# Strip superfluous subsecond values as Gerrit always just
# reports .000000000 for them anyway
date = date.split('.')[0]
# Pass back an invalid result if it falls after the requested
# cutoff
if before and date >= before:
return None
# Sanity check for completeness, but since "after" is also used
# in the Gerrit query this shouldn't ever actually be reached
if after and date < after:
return None
return date
def requester(url, params={}, headers={}):
"""A requests wrapper to consistently retry HTTPS queries"""
# Try up to 3 times
retry = requests.Session()
retry.mount("https://", requests.adapters.HTTPAdapter(max_retries=3))
return retry.get(url=url, params=params, headers=headers)
def decode_json(raw):
"""Trap JSON decoding failures and provide more detailed errors"""
# Gerrit's REST API prepends a JSON-breaker to avoid XSS vulnerabilities
if raw.text.startswith(")]}'"):
trimmed = raw.text[4:]
else:
trimmed = raw.text
# Try to decode and bail with much detail if it fails
try:
decoded = json.loads(trimmed)
except:
print('\nrequest returned %s error to query:\n\n %s\n'
'\nwith detail:\n\n %s\n' % (raw, raw.url, trimmed),
file=sys.stderr)
raise
return decoded
def query_gerrit(method, params={}):
"""Query the Gerrit REST API"""
# The base URL to Gerrit, for accessing both GitWeb and the REST
# API
GERRIT_API_URL = 'https://review.openstack.org/'
raw = requester(GERRIT_API_URL + method, params=params,
headers={'Accept': 'application/json'})
# Gitweb queries don't return JSON, so just pass them along as-is
if method is 'gitweb':
raw.encoding = 'utf-8' # Workaround for Gitweb encoding
return yaml.safe_load(raw.text)
else:
return decode_json(raw)
def lookup_member(email):
"""A requests wrapper to querying the OSF member directory API"""
# The OpenStack foundation member directory lookup API endpoint
MEMBER_LOOKUP_URL = 'https://openstackid-resources.openstack.org/'
# URL pattern for querying foundation members by E-mail address
raw = requester(
MEMBER_LOOKUP_URL + '/api/public/v1/members',
params={'filter[]': [
'group_slug==foundation-members',
'email==' + email,
]},
headers={'Accept': 'application/json'},
)
return decode_json(raw)
def usage(argv):
"""Parse command line argument"""
parser = argparse.ArgumentParser(
description="When run using OpenStack's Gerrit server, this builds "
"YAML representations of aggregate change owner details and change "
"counts for each governance project-team, as well as a combined set "
"for all teams. Before and after dates/times should be supplied in "
"formats Gerrit accepts: https://review.openstack.org/Documentation/"
"user-search.html#search-operators")
parser.add_argument("-a", "--after", help="Start date for matching merges")
parser.add_argument("-b", "--before", help="End date for matching merges")
parser.add_argument("-c", "--config", help="Path to script configuration")
parser.add_argument("-i", "--ignore", help="Account Id numbers to skip",
action='append')
parser.add_argument("-n", "--no-extra-atcs", help='Omit "extra ATCs"',
dest='no_extra_atcs', action='store_true')
parser.add_argument("-o", "--outdir", help="Create an output directory")
parser.add_argument("-r", "--ref", help="Specify a Governance refname")
parser.add_argument("-s", "--sieve", help="Add Gerrit query parameters")
return parser.parse_args(argv[1:])
def main(argv=sys.argv):
"""The giant pile of spaghetti which does everything else"""
# Record the start time for use later
start = datetime.datetime.utcnow()
options = usage(argv)
# If we're supplied a configuration file, use it
if options.config:
config = yaml.safe_load(open(options.config))
# Otherwise, use nothing
else:
config = {}
# Start of the match timeframe for change merges
if options.after:
after = options.after
elif 'after' in config:
after = config['after']
else:
after = None
# End of the match timeframe for change merges
if options.before:
before = options.before
elif 'before' in config:
before = config['before']
else:
before = None
# Owner Ids for whom to ignore changes
if options.ignore:
ignore = [int(i) for i in options.ignore]
elif 'ignore' in config:
ignore = config['ignore']
else:
ignore = []
# Whether to omit "extra ATCs"
if options.no_extra_atcs:
no_extra_atcs = options.no_extra_atcs
elif 'no-extra-atcs' in config:
no_extra_atcs = config['no-extra-atcs']
else:
no_extra_atcs = False
# Output file directory
if options.outdir:
outdir = options.outdir
elif 'outdir' in config:
outdir = config['outdir']
else:
outdir = '.'
if not os.path.isdir(outdir):
os.makedirs(outdir)
# Governance Git repository ref object for reference lists
if options.ref:
ref = options.ref
elif 'ref' in config:
ref = config['ref']
else:
ref = 'refs/heads/master'
# Gerrit change query additions
if options.sieve:
sieve = options.sieve
elif 'sieve' in config:
sieve = config['sieve']
else:
sieve = None
# The query identifying relevant changes
match = 'status:merged'
if after:
match = '%s after:"%s"' % (match, after)
if sieve:
match = '%s %s' % (match, sieve)
# Retrieve the governance projects list, needs a Git refname as a
# parameter
# TODO(fungi): make this a configurable option so that you can
# for example supply a custom project list for running elections
# in unofficial teams
gov_projects = query_gerrit(
'gitweb', {
'p': 'openstack/governance.git',
'a': 'blob_plain',
'f': 'reference/projects.yaml',
'hb': ref,
})
# A mapping of short (no prefix) to full repo names existing in
# Gerrit, used to handle repos which have a different namespace
# in governance during transitions and also to filter out repos
# listed in governance which don't actually exist
ger_repos = dict(
[(x.split('/')[-1], x) for x in query_gerrit('projects/')])
# This will be populated with change owners mapped to the
# project-teams maintaining their respective Git repositories
projects = {}
# This will be populated with all change owners and their
# account details
owners = {}
# This will be populated with discovered duplicate owners
duplicates = {}
# This will be populated with all individual E-mail addresses of
# change owners, to facilitate finding and merging duplicate
# accounts
all_emails = {}
# Iterate over all governance project-teams only at filename
# generation time
for project in gov_projects:
# This will be populated with change owner Ids and counts
projects[project] = {}
# Governance project-teams have one or more deliverables
for deliverable in gov_projects[project]['deliverables']:
# Each deliverable can have multiple repos
repos = gov_projects[project]['deliverables'][deliverable]['repos']
# Operate on repo short-names (no namespace) to avoid
# potential namespace mismatches between governance
# and Gerrit
for repo in [r.split('/')[-1] for r in repos]:
# Only process repos which actually exist in Gerrit,
# otherwise spew a warning if skipping
if repo not in ger_repos:
print('MISSING: %s' % repo, file=sys.stderr)
else:
# Query for an arbitrary change set and get
# detailed account information about the most
# recent patchset, paginating at 100 changes
offset = 0
changes = []
while offset >= 0:
changes += query_gerrit('changes/', params={
'q': 'project:%s %s' % (ger_repos[repo], match),
'n': '100',
'start': offset,
'o': [
'CURRENT_COMMIT',
'CURRENT_REVISION',
'DETAILED_ACCOUNTS',
],
})
if changes and changes[-1].get('_more_changes', False):
offset += 100
else:
offset = -1
# Iterate over each matched change in the repo
for change in changes:
# Get the merge date and skip if it's
# outside any requested date range
merged = date_merged(change, after, before)
if not merged:
continue
# We index owners by their unique Gerrit
# account Id numbers
owner = change['owner']['_account_id']
# If this owner is in the blacklist of Ids
# to skip, then move on to the next change
if owner in ignore:
continue
# Seen this owner already?
new_owner = owner
new = False
if owner in duplicates:
owner = duplicates[owner]
elif owner not in owners:
new = True
# For new additions, initialize this as
# their first and record specific account
# details
if new:
# Get the set of all E-mail addresses
# Gerrit knows for this owner's account
emails = query_gerrit(
'accounts/%s/emails'
% change['owner']['_account_id'])
# Find duplicate addresses and merge
# accounts when that happens
for email in emails:
address = normalize_email(email['email'])
if address in all_emails:
owner = all_emails[address]
duplicates[new_owner] = owner
print(
'MERGING DUPLICATE ACCOUNT: %s into %s'
% (new_owner, owner), file=sys.stderr)
break
# For newly found non-duplicate owners,
# initialize the global change count,
# newest/oldest merged dates, and an empty
# list where extra E-mail addresses can be
# added; also track their full name and
# Gerrit username
if new and owner == new_owner:
# TODO(fungi): this is a prime candidate
# to become a struct, or maybe a class
owners[owner] = {
'count': 1,
'extra': [],
'name': change['owner'].get('name'),
'newest': merged,
'oldest': merged,
'username': change['owner'].get('username'),
}
# If we've seen this owner on another change
# in any repo then just iterate their global
# change counter and update newest/oldest
# dates
else:
owners[owner]['count'] += 1
if merged > owners[owner]['newest']:
owners[owner]['newest'] = merged
elif merged < owners[owner]['oldest']:
owners[owner]['oldest'] = merged
# We only want to add addresses if this is a
# new owner or a new duplicate
if new:
# Iterate over each E-mail address
for email in emails:
# Normalize the address before
# performing any matching since
# Gerrit doesn't do a great job of
# this on its own
address = normalize_email(email['email'])
# Track this in the full list of all
# known E-mail addresses
all_emails[address] = owner
# Whether Gerrit considers this the
# preferred E-mail address
preferred = email.get('preferred', False)
# Store the preferred E-mail address
# under its own key since it has a
# special status, but only if this
# is not a duplicate account
if preferred and owner == new_owner:
owners[owner]['preferred'] = address
# If this was already added to
# the extras list due to an
# additional pre-normalized
# copy, remove it there
if address in owners[owner]['extra']:
owners[owner]['extra'].remove(address)
# Store a list of non-preferred
# addresses, deduplicating them in
# case they match post-normalization
# and treating duplicate preferred
# addresses as # non-preferred
else:
if ((address not in owners[owner]['extra'])
and (address != owners[owner].get(
'preferred', ''))):
owners[owner]['extra'].append(address)
# If we've seen this owner on another change
# in a repo under this project-team then
# just iterate their team change counter and
# update newest/oldest dates
if owner in projects[project]:
projects[project][owner]['count'] += 1
if merged > projects[project][owner]['newest']:
projects[project][owner]['newest'] = merged
elif merged < projects[project][owner]['oldest']:
projects[project][owner]['oldest'] = merged
# ...otherwise initialize this as their
# first
else:
# TODO(fungi): another potential struct
projects[project][owner] = {
'count': 1,
'newest': merged,
'oldest': merged,
}
# The negative counter will be used as a makeshift account Id
# for non-code contributors; those with owned changes use their
# Gerrit account Id instead
counter = 1
# Use the before time as the only contribution time for non-code
# contributors, falling back on the script start time if before
# was not specified
if before:
if len(before) == 10:
stamp = before + ' 00:00:00'
else:
stamp = before
else:
stamp = start.isoformat(sep=' ').split('.')[0]
# Iterate over all extra-atcs entries
if not no_extra_atcs:
for project in gov_projects:
for extra_atc in gov_projects[project].get('extra-atcs', []):
name = extra_atc['name']
email = extra_atc['email']
address = normalize_email(email)
if address in all_emails:
owner = all_emails[address]
else:
owner = -counter
all_emails[address] = owner
owners[owner] = {
'count': -1,
'extra': [],
'name': name,
'newest': stamp,
'oldest': stamp,
'preferred': address,
'username': '_non_code_contributor',
}
if owner not in projects[project]:
projects[project][owner] = {
'count': -1,
'newest': stamp,
'oldest': stamp,
}
counter += 1
# This will hold an address list for TC electorate rolls
electorate = []
# A table of owners for summit invites
invites = []
# A fresh pass through the owners to build some other datasets
for owner in owners:
# Sort extra E-mail address lists for ease of comparison
owners[owner]['extra'].sort()
# Build the data used for an invite
invite = []
if 'username' not in owners[owner] or not owners[owner]['username']:
print(
'SKIPPING MALFORMED OWNER: no username found for account %s' %
owner, file=sys.stderr)
continue
invite.append(owners[owner]['username'])
if 'name' not in owners[owner] or not owners[owner]['name']:
print(
'SKIPPING MALFORMED OWNER: no fullname found for account %s' %
owner, file=sys.stderr)
continue
invite.append(owners[owner]['name'].encode('utf-8'))
if 'preferred' not in owners[owner]:
if 'extra' in owners[owner] and owners[owner]['extra']:
owners[owner]['preferred'] = owners[owner]['extra'][0]
owners[owner]['extra'] = owners[owner]['extra'][1:]
print(
'MISSING PREFERRED EMAIL: used first extra address as '
'account %s preferred' % owner, file=sys.stderr)
else:
print(
'SKIPPING MALFORMED OWNER: no preferred or extra '
'addresses found for account %s' % owner, file=sys.stderr)
continue
for email in [owners[owner]['preferred']] + owners[owner]['extra']:
member = lookup_member(email)
if member['data']:
owners[owner]['member'] = member['data'][0]['id']
continue
invite.append(owners[owner]['preferred'])
invite += owners[owner]['extra']
invites.append(invite)
# Append preferred addresses to the TC electorate for members only
if 'member' in owners[owner]:
electorate.append(owners[owner]['preferred'] + '\n')
# Write out a YAML file covering all change owners
fd = open(os.path.join(outdir, '_all_owners.yaml'), 'w')
dumper(owners, stream=fd)
fd.close()
# Write out a YAML file covering tracked duplicate accounts
fd = open(os.path.join(outdir, '_duplicate_owners.yaml'), 'w')
dumper(duplicates, stream=fd)
fd.close()
# Write out a team-specific electoral roll for CIVS
fd = open(os.path.join(outdir, '_electorate.txt'), 'w')
fd.writelines(electorate)
fd.close()
# Write out a CSV file appropriate for the invite2summit tool
fd = open(os.path.join(outdir, '_invites.csv'), 'w')
csv.writer(fd).writerows(invites)
fd.close()
# Make another pass through the projects so they can be dumped
# to our output files
for project in projects:
# This will hold team-specific info for writing
output = {}
# This will hold an address list for PTL electoral rolls
electorate = []
# Use a normalized project name for output file names
normalized_project = normalize_project(project)
# Iterate over each change owner for the current team
for owner in projects[project]:
# Copy the global owner details into our output since
# we're going to modify some
output[owner] = dict(owners[owner])
# Replace the owner change count and newest/oldest
# merged dates with the team-specific value rather than
# using the count from the global set
for field in ('count', 'newest', 'oldest'):
output[owner][field] = projects[project][owner][field]
# Append preferred member addresses to the PTL electoral rolls
if 'member' in owners[owner]:
electorate.append(owners[owner]['preferred'] + '\n')
# Write out a team-specific YAML file
fd = open(os.path.join(outdir, '%s.yaml' % normalized_project), 'w')
dumper(output, stream=fd)
fd.close()
# Write out a team-specific electoral roll for CIVS
fd = open(os.path.join(outdir, '%s.txt' % normalized_project), 'w')
fd.writelines(electorate)
fd.close()
if __name__ == "__main__":
main()