获取云硬盘列表bug

有段时间没写博客了，主要还是刚进新公司，多少有点不适应（真会给自己找理由，明明是手里没技术，肚子里没货，能写啥！）。前面几个星期都在修一修horizon的bug，这个没啥很大难度，就是用了一个django的框架，然后再嵌套一些js框架，简明易懂。不扯远了，来点干货。。

浏览器抓包可以看到，这样一个请求耗时2.4秒，再把日志打到horizon代码中，看下结果如何：

class GetVolumes(APIMeta):

    def handle(self):
        status = self.request.POST.get('status', '')
        tenant_id = self.request.POST.get('tenant_id', '')
        data = []
        instances_info = {}
        LOG.error("step1:%s" % time.time())
        # import pdb
        # pdb.set_trace()
        volumes_obj = api.cinder.volume_list(self.request,
                                             {'all_tenants': 1})
        LOG.error("step2:%s" % time.time())
        volumes = filters.add_tenant_info(self.request,
                              volumes_obj,
                              "os-vol-tenant-attr:tenant_id")
        LOG.error("step3:%s" % time.time())
        instances = api.nova.server_list(self.request, all_tenants=True)
        LOG.error("step4:%s" % time.time())
        for ins in instances[0]:
            instances_info[ins.id] = ins.name
        if status:
            volumes = filter(lambda x: status in x.status, volumes)
        if tenant_id:
            volumes = filter(lambda x: \
                    getattr(x, 'os-vol-tenant-attr:tenant_id') == tenant_id,
                    volumes)
        LOG.error("step5:%s" % time.time())
        for volume in volumes:
            if volume.attachments:
                instance_id = volume.attachments[0]['server_id']
                instance_name = instances_info.get(instance_id,
                                                   instance_id)
            else:
                instance_name = ''
            tenant = getattr(volume, 'tenant', '')
            data.append({'volume_id': volume.id,
                         'display_name': volume.display_name,
                         'display_description': volume.display_description,
                         'bootable': volume.bootable,
                         'size': volume.size,
                         'status': volume.status,
                         'volume_type': volume.volume_type,
                         'host': getattr(volume,
                             "os-vol-host-attr:host", None),
                         'tenant_name': getattr(tenant, "name", None),
                         'tenant_id': getattr(tenant, "id", None),
                         'instance_name': instance_name,
                         })
        LOG.error("step6:%s" % time.time())
        return sorted(data, key = lambda x: x['display_name'], reverse = False)

测试结果如下：

-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step1:1430963063.15
-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step2:1430963064.71
-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step3:1430963064.8
-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step4:1430963066.29
-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step5:1430963066.29
-- ::, openstack_dashboard.dashboards.admin.volumes.json_views[line:] ERROR step6:1430963066.29

可以看到，主要的耗时在step1-step2 step3-step4时间，定位到相应代码段：

# step1与step2之间耗时：
volumes_obj = api.cinder.volume_list(self.request,{'all_tenants': 1})
# step3与step4之间耗时：
instances = api.nova.server_list(self.request, all_tenants=True)

再在cinderclient做时间戳：

def volume_list(request, search_opts=None):
    """
    To see all volumes in the cloud as an admin you can pass in a special
    search option: {'all_tenants': 1}
    """
    c_client = cinderclient(request)
    if c_client is None:
        return []
    import time
    LOG.error("horizon_start time: %s " % time.time())
    result = c_client.volumes.list(search_opts=search_opts)
    LOG.error("horizon_end time: %s " % time.time())
    return result
#    return c_client.volumes.list(search_opts=search_opts)

结果如下：

-- ::, openstack_dashboard.api.cinder[line:] ERROR horizon_start time: 1431069736.4
-- ::, openstack_dashboard.api.cinder[line:] ERROR horizon_end time: 1431069739.85 

再在cinder服务端设置时间戳：/cinder/api/v1/volumes.py

    def _items(self, req, entity_maker):
        # import pdb
        # pdb.set_trace()
        """Returns a list of volumes, transformed through entity_maker."""

        #pop out limit and offset , they are not search_opts
        LOG.debug(_("time1------------------------%s" % time.time()))
        search_opts = req.GET.copy()
        search_opts.pop('limit', None)
        search_opts.pop('offset', None)

        if 'metadata' in search_opts:
            search_opts['metadata'] = ast.literal_eval(search_opts['metadata'])

        context = req.environ['cinder.context']
        remove_invalid_options(context,
                               search_opts, self._get_volume_search_options())

        volumes = self.volume_api.get_all(context, marker=None, limit=None,
                                          sort_key='created_at',
                                          sort_dir='desc', filters=search_opts)

        volumes = [dict(vol.iteritems()) for vol in volumes]
        LOG.debug(_("time2------------------------%s" % time.time()))

        for volume in volumes:
            self._add_visible_admin_metadata(context, volume)

        limited_list = common.limited(volumes, req)
        LOG.debug(_("time3------------------------%s" % time.time()))
        req.cache_resource(limited_list)
        res = [entity_maker(context, vol) for vol in limited_list]
        LOG.debug(_("time4------------------------%s" % time.time()))
        return {'volumes': res}

结果如下：

time1------------------------1431069737.57
time2------------------------1431069738.02
time3------------------------1431069738.02
time4------------------------1431069738.05

显而易见，时间都消耗在cinder服务端向cinder客户端返回数据上面了，下面分析一下返回的数据：

{'migration_status': None,
'availability_zone': u'nova',
'terminated_at': None,
'updated_at': datetime.datetime(, , , , , ),
'provider_geometry': None,
'snapshot_id': None,
'ec2_id': None,
'mountpoint': None,
'deleted_at': None,
'id': u'69400200-6dea-41ec-ab7f-e3719fc9b18f',
'size': 3L,
'user_id': u'33d5edff940f4812b5368e54499991e4',
'attach_time': None,
'attached_host': None,
'display_description': u'',
'encryption_key_id': None,
'project_id': u'01f2a2e8594342d78865376cd81e2c67',
'launched_at': datetime.datetime(, , , , , ),
'scheduled_at': datetime.datetime(, , , , , ),
'status': u'available',
'volume_type_id': u'3a255382-1da3-4449-ad5b-5f37a4cf3d2c',
'deleted': False,
'provider_location': None,
'host': u'test1',
'source_volid': None,
'provider_auth': None,
'display_name': u'disk-102',
'instance_uuid': None,
'bootable': False,
'created_at': datetime.datetime(, , , , , ),
'attach_status': u'detached',
'volume_type': <cinder.db.sqlalchemy.models.VolumeTypes object at 0x44277d0>,
'_name_id': None, 'volume_metadata': []}, 

上面字段基本都会用到，看来解决方案不能在精简数据上面了，或者可以在cinder客户端做一个缓存，未完待续。。