quay/buildman/jobutil/buildjob.py

import json

from cachetools import lru_cache
from endpoints.notificationhelper import spawn_notification
from data import model


class BuildJobLoadException(Exception):
  """ Exception raised if a build job could not be instantiated for some reason. """
  pass

class BuildJob(object):
  """ Represents a single in-progress build job. """
  def __init__(self, job_item):
    self.job_item = job_item

    try:
      self.job_details = json.loads(job_item.body)
    except ValueError:
      raise BuildJobLoadException(
          'Could not parse build queue item config with ID %s' % self.job_details['build_uuid']
      )

  def send_notification(self, kind, error_message=None):
    tags = self.build_config.get('docker_tags', ['latest'])
    event_data = {
      'build_id': self.repo_build.uuid,
      'build_name': self.repo_build.display_name,
      'docker_tags': tags,
      'trigger_id': self.repo_build.trigger.uuid,
      'trigger_kind': self.repo_build.trigger.service.name
    }

    if error_message is not None:
      event_data['error_message'] = error_message

    spawn_notification(self.repo_build.repository, kind, event_data,
                       subpage='build?current=%s' % self.repo_build.uuid,
                       pathargs=['build', self.repo_build.uuid])


  @lru_cache(maxsize=1)
  def _load_repo_build(self):
    try:
      return model.get_repository_build(self.job_details['build_uuid'])
    except model.InvalidRepositoryBuildException:
      raise BuildJobLoadException(
          'Could not load repository build with ID %s' % self.job_details['build_uuid'])

  @property
  def repo_build(self):
    return self._load_repo_build()

  @property
  def pull_credentials(self):
    """ Returns the pull credentials for this job, or None if none. """
    return self.job_details.get('pull_credentials')

  @property
  def build_config(self):
    try:
      return json.loads(self.repo_build.job_config)
    except ValueError:
      raise BuildJobLoadException(
          'Could not parse repository build job config with ID %s' % self.job_details['build_uuid']
      )

  def determine_cached_tag(self, base_image_id=None, cache_comments=None):
    """ Returns the tag to pull to prime the cache or None if none. """
    cached_tag = None
    if base_image_id and cache_comments:
      cached_tag = self._determine_cached_tag_by_comments(base_image_id, cache_comments)

    if not cached_tag:
      cached_tag = self._determine_cached_tag_by_tag()

    return cached_tag

  def _determine_cached_tag_by_comments(self, base_image_id, cache_commands):
    """ Determines the tag to use for priming the cache for this build job, by matching commands
        starting at the given base_image_id. This mimics the Docker cache checking, so it should,
        in theory, provide "perfect" caching.
    """
    # Lookup the base image in the repository. If it doesn't exist, nothing more to do.
    repo_namespace = self._repo_build.repository.namespace_user.username
    repo_name = self._repo_build.repository.name

    repository = model.get_repository(repo_namespace, repo_name)
    if repository is None:
      # Should never happen, but just to be sure.
      return None

    current_image = model.get_image(repository, base_image_id)
    if current_image is None:
      return None

    # For each cache comment, find a child image that matches the command.
    for cache_command in cache_commands:
      full_command = '["/bin/sh", "-c", "%s"]' % cache_command
      current_image = model.find_child_image(repository, current_image, full_command)
      if current_image is None:
        return None

    # Find a tag associated with the image, if any.
    # TODO(jschorr): We should just return the image ID instead of a parent tag, OR we should
    # make this more efficient.
    for tag in model.list_repository_tags(repo_namespace, repo_name):
      tag_image = tag.image
      ancestor_index = '/%s/' % current_image.id
      if ancestor_index in tag_image.ancestors:
        return tag.name

    return None

  def _determine_cached_tag_by_tag(self):
    """ Determines the cached tag by looking for one of the tags being built, and seeing if it
        exists in the repository. This is a fallback for when no comment information is available.
    """
    tags = self._build_config.get('docker_tags', ['latest'])
    existing_tags = model.list_repository_tags(self._repo_build.repository.namespace_user.username,
                                               self._repo_build.repository.name)

    cached_tags = set(tags) & set([tag.name for tag in existing_tags])
    if cached_tags:
      return list(cached_tags)[0]

    return None