view rhodecode/lib/celerylib/tasks.py @ 4116:ffd45b185016 rhodecode-2.2.5-gpl

Imported some of the GPLv3'd changes from RhodeCode v2.2.5. This imports changes between changesets 21af6c4eab3d and 6177597791c2 in RhodeCode's original repository, including only changes to Python files and HTML. RhodeCode clearly licensed its changes to these files under GPLv3 in their /LICENSE file, which states the following: The Python code and integrated HTML are licensed under the GPLv3 license. (See: https://code.rhodecode.com/rhodecode/files/v2.2.5/LICENSE or http://web.archive.org/web/20140512193334/https://code.rhodecode.com/rhodecode/files/f3b123159901f15426d18e3dc395e8369f70ebe0/LICENSE for an online copy of that LICENSE file) Conservancy reviewed these changes and confirmed that they can be licensed as a whole to the Kallithea project under GPLv3-only. While some of the contents committed herein are clearly licensed GPLv3-or-later, on the whole we must assume the are GPLv3-only, since the statement above from RhodeCode indicates that they intend GPLv3-only as their license, per GPLv3ยง14 and other relevant sections of GPLv3.
author Bradley M. Kuhn <bkuhn@sfconservancy.org>
date Wed, 02 Jul 2014 19:03:13 -0400
parents 3563bb7b4b82
children 7e5f8c12a3fc
line wrap: on
line source

# -*- coding: utf-8 -*-
# This program is free software: you can redistribute it and/or modify
# it under the terms of the GNU General Public License as published by
# the Free Software Foundation, either version 3 of the License, or
# (at your option) any later version.
#
# This program is distributed in the hope that it will be useful,
# but WITHOUT ANY WARRANTY; without even the implied warranty of
# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
# GNU General Public License for more details.
#
# You should have received a copy of the GNU General Public License
# along with this program.  If not, see <http://www.gnu.org/licenses/>.
"""
rhodecode.lib.celerylib.tasks
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

RhodeCode task modules, containing all task that suppose to be run
by celery daemon

:created_on: Oct 6, 2010
:author: marcink
:copyright: (c) 2013 RhodeCode GmbH.
:license: GPLv3, see LICENSE for more details.
"""

from celery.decorators import task

import os
import traceback
import logging
from os.path import join as jn

from time import mktime
from operator import itemgetter
from string import lower

from pylons import config, url
from pylons.i18n.translation import _

from rhodecode.lib.vcs import get_backend

from rhodecode import CELERY_ON, CELERY_EAGER
from rhodecode.lib.utils2 import safe_str
from rhodecode.lib.celerylib import run_task, locked_task, dbsession, \
    str2bool, __get_lockkey, LockHeld, DaemonLock, get_session
from rhodecode.lib.helpers import person
from rhodecode.lib.rcmail.smtp_mailer import SmtpMailer
from rhodecode.lib.utils import add_cache, action_logger
from rhodecode.lib.compat import json, OrderedDict
from rhodecode.lib.hooks import log_create_repository

from rhodecode.model.db import Statistics, Repository, User
from rhodecode.model.scm import ScmModel


add_cache(config)  # pragma: no cover

__all__ = ['whoosh_index', 'get_commits_stats',
           'reset_user_password', 'send_email']


def get_logger(cls):
    if CELERY_ON:
        try:
            log = cls.get_logger()
        except Exception:
            log = logging.getLogger(__name__)
    else:
        log = logging.getLogger(__name__)

    return log


@task(ignore_result=True)
@locked_task
@dbsession
def whoosh_index(repo_location, full_index):
    from rhodecode.lib.indexers.daemon import WhooshIndexingDaemon
    log = get_logger(whoosh_index)
    DBS = get_session()

    index_location = config['index_dir']
    WhooshIndexingDaemon(index_location=index_location,
                         repo_location=repo_location, sa=DBS)\
                         .run(full_index=full_index)


@task(ignore_result=True)
@dbsession
def get_commits_stats(repo_name, ts_min_y, ts_max_y, recurse_limit=100):
    log = get_logger(get_commits_stats)
    DBS = get_session()
    lockkey = __get_lockkey('get_commits_stats', repo_name, ts_min_y,
                            ts_max_y)
    lockkey_path = config['app_conf']['cache_dir']

    log.info('running task with lockkey %s' % lockkey)

    try:
        lock = l = DaemonLock(file_=jn(lockkey_path, lockkey))

        # for js data compatibility cleans the key for person from '
        akc = lambda k: person(k).replace('"', "")

        co_day_auth_aggr = {}
        commits_by_day_aggregate = {}
        repo = Repository.get_by_repo_name(repo_name)
        if repo is None:
            return True

        repo = repo.scm_instance
        repo_size = repo.count()
        # return if repo have no revisions
        if repo_size < 1:
            lock.release()
            return True

        skip_date_limit = True
        parse_limit = int(config['app_conf'].get('commit_parse_limit'))
        last_rev = None
        last_cs = None
        timegetter = itemgetter('time')

        dbrepo = DBS.query(Repository)\
            .filter(Repository.repo_name == repo_name).scalar()
        cur_stats = DBS.query(Statistics)\
            .filter(Statistics.repository == dbrepo).scalar()

        if cur_stats is not None:
            last_rev = cur_stats.stat_on_revision

        if last_rev == repo.get_changeset().revision and repo_size > 1:
            # pass silently without any work if we're not on first revision or
            # current state of parsing revision(from db marker) is the
            # last revision
            lock.release()
            return True

        if cur_stats:
            commits_by_day_aggregate = OrderedDict(json.loads(
                                        cur_stats.commit_activity_combined))
            co_day_auth_aggr = json.loads(cur_stats.commit_activity)

        log.debug('starting parsing %s' % parse_limit)
        lmktime = mktime

        last_rev = last_rev + 1 if last_rev >= 0 else 0
        log.debug('Getting revisions from %s to %s' % (
             last_rev, last_rev + parse_limit)
        )
        for cs in repo[last_rev:last_rev + parse_limit]:
            log.debug('parsing %s' % cs)
            last_cs = cs  # remember last parsed changeset
            k = lmktime([cs.date.timetuple()[0], cs.date.timetuple()[1],
                          cs.date.timetuple()[2], 0, 0, 0, 0, 0, 0])

            if akc(cs.author) in co_day_auth_aggr:
                try:
                    l = [timegetter(x) for x in
                         co_day_auth_aggr[akc(cs.author)]['data']]
                    time_pos = l.index(k)
                except ValueError:
                    time_pos = None

                if time_pos >= 0 and time_pos is not None:

                    datadict = \
                        co_day_auth_aggr[akc(cs.author)]['data'][time_pos]

                    datadict["commits"] += 1
                    datadict["added"] += len(cs.added)
                    datadict["changed"] += len(cs.changed)
                    datadict["removed"] += len(cs.removed)

                else:
                    if k >= ts_min_y and k <= ts_max_y or skip_date_limit:

                        datadict = {"time": k,
                                    "commits": 1,
                                    "added": len(cs.added),
                                    "changed": len(cs.changed),
                                    "removed": len(cs.removed),
                                   }
                        co_day_auth_aggr[akc(cs.author)]['data']\
                            .append(datadict)

            else:
                if k >= ts_min_y and k <= ts_max_y or skip_date_limit:
                    co_day_auth_aggr[akc(cs.author)] = {
                                        "label": akc(cs.author),
                                        "data": [{"time":k,
                                                 "commits":1,
                                                 "added":len(cs.added),
                                                 "changed":len(cs.changed),
                                                 "removed":len(cs.removed),
                                                 }],
                                        "schema": ["commits"],
                                        }

            #gather all data by day
            if k in commits_by_day_aggregate:
                commits_by_day_aggregate[k] += 1
            else:
                commits_by_day_aggregate[k] = 1

        overview_data = sorted(commits_by_day_aggregate.items(),
                               key=itemgetter(0))

        if not co_day_auth_aggr:
            co_day_auth_aggr[akc(repo.contact)] = {
                "label": akc(repo.contact),
                "data": [0, 1],
                "schema": ["commits"],
            }

        stats = cur_stats if cur_stats else Statistics()
        stats.commit_activity = json.dumps(co_day_auth_aggr)
        stats.commit_activity_combined = json.dumps(overview_data)

        log.debug('last revison %s' % last_rev)
        leftovers = len(repo.revisions[last_rev:])
        log.debug('revisions to parse %s' % leftovers)

        if last_rev == 0 or leftovers < parse_limit:
            log.debug('getting code trending stats')
            stats.languages = json.dumps(__get_codes_stats(repo_name))

        try:
            stats.repository = dbrepo
            stats.stat_on_revision = last_cs.revision if last_cs else 0
            DBS.add(stats)
            DBS.commit()
        except:
            log.error(traceback.format_exc())
            DBS.rollback()
            lock.release()
            return False

        # final release
        lock.release()

        # execute another task if celery is enabled
        if len(repo.revisions) > 1 and CELERY_ON and recurse_limit > 0:
            recurse_limit -= 1
            run_task(get_commits_stats, repo_name, ts_min_y, ts_max_y,
                     recurse_limit)
        if recurse_limit <= 0:
            log.debug('Breaking recursive mode due to reach of recurse limit')
        return True
    except LockHeld:
        log.info('LockHeld')
        return 'Task with key %s already running' % lockkey


@task(ignore_result=True)
@dbsession
def send_email(recipients, subject, body='', html_body=''):
    """
    Sends an email with defined parameters from the .ini files.

    :param recipients: list of recipients, it this is empty the defined email
        address from field 'email_to' is used instead
    :param subject: subject of the mail
    :param body: body of the mail
    :param html_body: html version of body
    """
    log = get_logger(send_email)
    DBS = get_session()

    email_config = config
    subject = "%s %s" % (email_config.get('email_prefix', ''), subject)
    if not recipients:
        # if recipients are not defined we send to email_config + all admins
        admins = [u.email for u in User.query()
                  .filter(User.admin == True).all()]
        recipients = [email_config.get('email_to')] + admins

    mail_from = email_config.get('app_email_from', 'RhodeCode')
    user = email_config.get('smtp_username')
    passwd = email_config.get('smtp_password')
    mail_server = email_config.get('smtp_server')
    mail_port = email_config.get('smtp_port')
    tls = str2bool(email_config.get('smtp_use_tls'))
    ssl = str2bool(email_config.get('smtp_use_ssl'))
    debug = str2bool(email_config.get('debug'))
    smtp_auth = email_config.get('smtp_auth')

    if not mail_server:
        log.error("SMTP mail server not configured - cannot send mail")
        return False

    try:
        m = SmtpMailer(mail_from, user, passwd, mail_server, smtp_auth,
                       mail_port, ssl, tls, debug=debug)
        m.send(recipients, subject, body, html_body)
    except:
        log.error('Mail sending failed')
        log.error(traceback.format_exc())
        return False
    return True

@task(ignore_result=False)
@dbsession
def create_repo(form_data, cur_user):
    from rhodecode.model.repo import RepoModel
    from rhodecode.model.user import UserModel
    from rhodecode.model.db import RhodeCodeSetting

    log = get_logger(create_repo)
    DBS = get_session()

    cur_user = UserModel(DBS)._get_user(cur_user)

    owner = cur_user
    repo_name = form_data['repo_name']
    repo_name_full = form_data['repo_name_full']
    repo_type = form_data['repo_type']
    description = form_data['repo_description']
    private = form_data['repo_private']
    clone_uri = form_data.get('clone_uri')
    repo_group = form_data['repo_group']
    landing_rev = form_data['repo_landing_rev']
    copy_fork_permissions = form_data.get('copy_permissions')
    copy_group_permissions = form_data.get('repo_copy_permissions')
    fork_of = form_data.get('fork_parent_id')
    state = form_data.get('repo_state', Repository.STATE_PENDING)

    # repo creation defaults, private and repo_type are filled in form
    defs = RhodeCodeSetting.get_default_repo_settings(strip_prefix=True)
    enable_statistics = defs.get('repo_enable_statistics')
    enable_locking = defs.get('repo_enable_locking')
    enable_downloads = defs.get('repo_enable_downloads')

    try:
        repo = RepoModel(DBS)._create_repo(
            repo_name=repo_name_full,
            repo_type=repo_type,
            description=description,
            owner=owner,
            private=private,
            clone_uri=clone_uri,
            repo_group=repo_group,
            landing_rev=landing_rev,
            fork_of=fork_of,
            copy_fork_permissions=copy_fork_permissions,
            copy_group_permissions=copy_group_permissions,
            enable_statistics=enable_statistics,
            enable_locking=enable_locking,
            enable_downloads=enable_downloads,
            state=state
        )

        action_logger(cur_user, 'user_created_repo',
                      form_data['repo_name_full'], '', DBS)

        DBS.commit()
        # now create this repo on Filesystem
        RepoModel(DBS)._create_filesystem_repo(
            repo_name=repo_name,
            repo_type=repo_type,
            repo_group=RepoModel(DBS)._get_repo_group(repo_group),
            clone_uri=clone_uri,
        )
        repo = Repository.get_by_repo_name(repo_name_full)
        log_create_repository(repo.get_dict(), created_by=owner.username)

        # update repo changeset caches initially
        repo.update_changeset_cache()

        # set new created state
        repo.set_state(Repository.STATE_CREATED)
        DBS.commit()
    except Exception, e:
        log.warning('Exception %s occured when forking repository, '
                    'doing cleanup...' % e)
        # rollback things manually !
        repo = Repository.get_by_repo_name(repo_name_full)
        if repo:
            Repository.delete(repo.repo_id)
            DBS.commit()
            RepoModel(DBS)._delete_filesystem_repo(repo)
        raise

    # it's an odd fix to make celery fail task when exception occurs
    def on_failure(self, *args, **kwargs):
        pass

    return True


@task(ignore_result=False)
@dbsession
def create_repo_fork(form_data, cur_user):
    """
    Creates a fork of repository using interval VCS methods

    :param form_data:
    :param cur_user:
    """
    from rhodecode.model.repo import RepoModel
    from rhodecode.model.user import UserModel

    log = get_logger(create_repo_fork)
    DBS = get_session()

    base_path = Repository.base_path()
    cur_user = UserModel(DBS)._get_user(cur_user)

    repo_name = form_data['repo_name']  # fork in this case
    repo_name_full = form_data['repo_name_full']

    repo_type = form_data['repo_type']
    owner = cur_user
    private = form_data['private']
    clone_uri = form_data.get('clone_uri')
    repo_group = form_data['repo_group']
    landing_rev = form_data['landing_rev']
    copy_fork_permissions = form_data.get('copy_permissions')

    try:
        fork_of = RepoModel(DBS)._get_repo(form_data.get('fork_parent_id'))

        fork_repo = RepoModel(DBS)._create_repo(
            repo_name=repo_name_full,
            repo_type=repo_type,
            description=form_data['description'],
            owner=owner,
            private=private,
            clone_uri=clone_uri,
            repo_group=repo_group,
            landing_rev=landing_rev,
            fork_of=fork_of,
            copy_fork_permissions=copy_fork_permissions
        )
        action_logger(cur_user, 'user_forked_repo:%s' % repo_name_full,
                      fork_of.repo_name, '', DBS)
        DBS.commit()

        update_after_clone = form_data['update_after_clone']
        source_repo_path = os.path.join(base_path, fork_of.repo_name)

        # now create this repo on Filesystem
        RepoModel(DBS)._create_filesystem_repo(
            repo_name=repo_name,
            repo_type=repo_type,
            repo_group=RepoModel(DBS)._get_repo_group(repo_group),
            clone_uri=source_repo_path,
        )
        repo = Repository.get_by_repo_name(repo_name_full)
        log_create_repository(repo.get_dict(), created_by=owner.username)

        # update repo changeset caches initially
        repo.update_changeset_cache()

        # set new created state
        repo.set_state(Repository.STATE_CREATED)
        DBS.commit()
    except Exception, e:
        log.warning('Exception %s occured when forking repository, '
                    'doing cleanup...' % e)
        #rollback things manually !
        repo = Repository.get_by_repo_name(repo_name_full)
        if repo:
            Repository.delete(repo.repo_id)
            DBS.commit()
            RepoModel(DBS)._delete_filesystem_repo(repo)
        raise

    # it's an odd fix to make celery fail task when exception occurs
    def on_failure(self, *args, **kwargs):
        pass

    return True


def __get_codes_stats(repo_name):
    from rhodecode.config.conf import LANGUAGES_EXTENSIONS_MAP
    repo = Repository.get_by_repo_name(repo_name).scm_instance

    tip = repo.get_changeset()
    code_stats = {}

    def aggregate(cs):
        for f in cs[2]:
            ext = lower(f.extension)
            if ext in LANGUAGES_EXTENSIONS_MAP.keys() and not f.is_binary:
                if ext in code_stats:
                    code_stats[ext] += 1
                else:
                    code_stats[ext] = 1

    map(aggregate, tip.walk('/'))

    return code_stats or {}